WIRELESS devices are usually powered by limited battery

Size: px
Start display at page:

Download "WIRELESS devices are usually powered by limited battery"

Transcription

1 1158 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 Energy-Efficient Wireless Packet Scheduling with Quality of Service Control Xiliang Zhong, Student Member, IEEE, and Cheng-Zhong Xu, Senior Member, IEEE Abstract In this paper, we study the problem of packet scheduling in a wireless environment with the objective of minimizing the average transmission energy expenditure under individual packet delay constraints. Most past studies assumed that the input arrivals followed a Poisson process or were statistically independent. However, traffic from a real source typically has strong time correlation. We model a packet scheduling and queuing system for a general input process in linear time-invariant systems. We propose an energy-efficient packet scheduling policy that takes the correlation into account. Meanwhile, a slower transmission rate implies that packets stay in the transmitter for a longer time, which may result in unexpected transmitter overload and buffer overflow. We derive the upper bounds of the maximum transmission rate under an overload probability and the upper bounds of the required buffer size under a packet drop rate. Simulation results show that the proposed scheduler improves up to 15 percent in energy savings compared with the policies that assume statistically independent input. Evaluation of the bounds in providing QoS control shows that both deadline misses and packet drops can be effectively bounded by a predefined constraint. Index Terms Packet scheduling, power control, QoS, wireless networks. Ç 1 INTRODUCTION WIRELESS devices are usually powered by limited battery resources. Reliable content delivery over a wireless channel is a major source of energy expenditure. It is essential to reduce the power consumption of the devices without performance degradation. The objective of this study is to strike a good trade-off between transmission power and delay constraints. It is known that power consumption between two points over a wireless channel is exponentially related to the information transmission rate [2]. A linear increase of transmission time can achieve super linear energy savings. However, applications are often delay sensitive. The transmission time cannot be arbitrarily long. Researchers have proposed different approaches to deal with the energy-delay trade-off. In [2], [5], and [19], the average queuing delay was considered as a constraint in energy minimization. In [9], [25], and [29], a single deadline for all packets was set so that all arrivals in time ½0;TÞ have to be transmitted before time T. Both average delay and deadline constraints provide a delay guarantee for a group of packets; the transmission delay of an individual packet can still be quite large. A more desirable constraint for delay-sensitive applications is the individual packet delay under which each packet transmission must finish before its deadline [3], [10], [19], [21], [28]. However, most existing work can only be applied to special types of packet arrival. For example, the input was assumed to be a Poisson process in [28], independent and identically distributed over each slot in [10], [19], a periodic task with constant interarrival. The authors are with the Department of Electrical and Computer Engineering, Wayne State University, 5050 Anthony Wayne Drive, Detroit, MI {xlzhong, czxu}@wayne.edu. Manuscript received 18 Feb. 2006; revised 2 Sept. 2006; accepted 27 Nov. 2006; published online 7 Feb For information on obtaining reprints of this article, please send to: tmc@computer.org, and reference IEEECS Log Number TMC Digital Object Identifier no /TMC time in [21], and deterministic with all timing information known offline in [3]. However, we cannot always have full knowledge of input arrivals during packet scheduling and a Poisson process is limited in characterizing real-world traffic. For example, it has been shown that, in the wireless LAN environment, packet arrivals are more bursty [25]. The assumption of statistically dependent packet arrivals may not be valid in reality either. Existing studies confirm the presence of long range dependence (LRD) and selfsimilarity for packets in an Ethernet LAN [26], packets from a variable bit rate (VBR) video [1]. In this work, we focus on packet scheduling with an individual deadline constraint for a general input process in an Additive White Gaussian Noise (AWGN) channel. We present an energyefficient scheduling policy to take into account the input autocorrelation. We demonstrate that the input autocorrelation plays an important role in determining the scheduling policy. Furthermore, we prove that a recent approach [10] is essentially a special case of our proposed policy for statistically independent input. We observe that, because of the online nature of the scheduler and a slower transmission rate for energy efficiency, an energy-aware policy may lead to unexpected overload or packet drops. As packet loss and error are not uncommon in wireless systems, the application layer is usually designed to be tolerant to a certain degree of QoS violation [18]. We thus provide a controllable QoS guarantee by investigating the relationship between the maximum transmission rate and an overload probability and the relationship between queue size and a packet drop rate. The relationships can be used offline for capacity configuration subject to a QoS constraint and online for QoS bounds with a fixed capacity. The rest of the paper is organized as follows: In Section 2, we introduce the system model and present the problem formulation. In Section 3, we propose a modeling of the scheduling and present an energy-efficient scheduling /07/$25.00 ß 2007 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS

2 ZHONG AND XU: ENERGY-EFFICIENT WIRELESS PACKET SCHEDULING WITH QUALITY OF SERVICE CONTROL 1159 Fig. 1. System model. policy. Relationships of transmitter capacities and QoS constraints are revealed in Section 4. Section 5 verifies the analytical results through simulation. Section 6 reviews related work. Section 7 concludes the article. 2 PROBLEM FORMULATION We consider a single user time-slotted system shown in Fig. 1. Let t represent the system scheduling epoch index. The number of input packets to the transmitter at time t is given by a stationary random process xðtþ. The packets are assumed to be uniform in size and the size is long enough for reliable communication close to the mutual information of the channel. Each packet is associated with a QoS level characterized by a delay constraint t d. The packets to be transmitted are first stored in a queue with size s max. The queue backlog length at the beginning of time t is sðtþ. The transmitter schedules packets out of the queue at a rate of rðtþ at time t and uses power PðrðtÞÞ for transmission. We consider a transmitter in support of multiple transmission rates. In practice, this can be achieved by using dynamic modulation scaling [20], coding scaling [25], or a change of symbol rate [4]. Consider a AWGN channel between a pair of transmitters and receivers. The maximum channel capacity under optimal channel coding is C ¼ 1 2 log 2 1 þ PðrðtÞÞ bits=transmission; N where N is noise power. Let B denote the channel bandwidth. We can represent the relationship between transmission rate rðtþ and transmission power PðrðtÞÞ as PðrðtÞÞ ¼ N 2 2rðtÞ B 1 : ð1þ The power function is monotonically increasing and strictly convex with respect to rðtþ. The same characteristics of power function apply even if suboptimal channel coding is deployed [25]. This means that even a small reduction in the transmission rate or increase in transmission delay can lead to a large energy savings. We consider an input arrival process during time period ½0;TÞ in an AWGN channel. The AWGN channel does not account for the phenomena of channel fading, interference, and dispersion. The average energy consumed by the input process can be expressed as IE½PðrðtÞÞŠ ¼ 1 T P T t¼0 PðrðtÞÞ. A scheduler is optimal if it leads to the minimum amount of average energy consumption. Such an optimal scheduler relies upon complete knowledge of future packet arrivals, which is not available in online scheduling for a general input process. It is known that IE½PðrðtÞÞŠ PðIE½rðtÞŠÞ due to the convexity of the energy function. That is, the maximum energy savings would be obtained if we operated the transmitter at the long-term average service rate. Because of the variation of input arrivals, a constant rate may not guarantee all packets transmitted before their deadlines. Considering the delay constraint, we expect an energy-efficient feasible scheduler to assign the transmission rate to the average service rate IE½rðtÞŠ as close as possible. We denote IE½rðtÞŠ as r. Our objective is to find an online scheduler that minimizes the mean quare errors between the transmission rate and the constant rate r, IE½ðrðtÞ r Þ 2 Š. 3 ENERGY-EFFICIENT SCHEDULING In this section, we first present a model of packet scheduling in a linear system. The model facilitates the derivation of an energy-efficient packet scheduling policy. 3.1 An Energy-Aware Scheduling Model To characterize the transmission process, we define a transmission function as the amount of packets transmitted during each time slot, Tðt; t a ;xðt a ÞÞ. It is a function of system time t, packet arrival time t a, and the number of packets arrived during time ½t a 1;t a Þ, xðt a Þ. A general scheduler should also consider relative deadline t d of each packet. Due to the convexity of the power function, any packet transmitted before its deadline t d can be delayed with reduced transmission rate and energy. We hence consider packet transmitted before its deadline to effectively complete at the deadline to achieve maximum energy savings. We define a scheduling function h r ðt; t a Þ as the portion of packets transmitted during time ½t; t þ 1Þ for packets arrived at time t a. The number of transmitted packets from xðt a Þ at the slot is then h r ðt; t a Þxðt a Þ. The total transmission at time t is a sum of all transmitted packets that arrive during the last t d time slots. That is, rðtþ ¼ Xt t a¼t t dþ1 Tðt; t a ;xðt a ÞÞ ¼ Xt t a¼t t dþ1 xðt a Þh r ðt; t a Þ: Equation (2) models the scheduling process as a singleinput single-output linear system and the scheduling function h r ðt; t a Þ is its unit impulse response. A scheduler is normally causal, which means no resource will be reserved before packet arrivals, i.e., for all t<t a, h r ðt; t a Þ¼0. We consider a time-invariant scheduler in which its impulse response does not depend on the time when the impulse is applied. Thus, if an impulse, occurring at t ¼ 0, causes the response h r ðtþ, then an impulse, occurring at t ¼ t a, must cause the response h r ðt t a Þ. This means h r ðt; t a Þ¼h r ðt t a Þ. The transmission rate can be expressed as rðtþ ¼ Xt t a¼t t dþ1 xðt a Þh r ðt t a Þ¼xðtÞh r ðtþ; ð2þ ð3þ

3 1160 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 where is a convolution operator. Equation (3) models the scheduling process as a perfect form in a linear, timeinvariant (LTI) system with a transfer function h r ðtþ. As 0 h r ðtþ 1, the system represented by h r ðtþ is a low pass filter. 3.2 An Energy-Efficient Scheduling Policy Consider packets that arrived at time t a, xðt a Þ. We try to transmit all the packets before deadline t a þ t d. The number of packets to be transmitted is t aþt xðt a Þ¼ X d 1 t a Tðt; t a ;xðt a ÞÞ ¼ X þt d 1 h r ðt; t a Þxðt a Þ t¼t a That is, taþtd 1 ¼ X t a Xþt d 1 t¼t a t¼t a h r ðt t a Þxðt a Þ: h r ðt t a Þ¼1 or t¼t a Xt d 1 t¼0 h r ðtþ ¼1: Define h ¼½h r ð0þ; h r ð1þ;...;h r ðt d 1ÞŠ 0 and xðtþ ¼ ½xðtÞ;xðt 1Þ;...;xðt t d þ 1ÞŠ as vector forms of the scheduling function and packet arrivals. Let ¼ IE½xðtÞx 0 ðtþš, being the covariance matrix of input process xðtþ in the order of t d. Consider a wide-sense stationary (WSS) input process, in which the mean of xðtþ is constant and its autocorrelation function depends only on the time difference. We note that the assumption of WSS is general in modeling self-similar traffic. For example, traffic traces generated from synthetic Fractional Gaussian Noise processes (FGN) are WSS. We present the solution to the optimization problem for a WSS process in the following theorem: Theorem 3.1. The optimal delay bounded scheduler for a WSS process has a scheduling function in the form of h ¼ 1 u 0 u 1 u 0 : Proof. Rewrite (3) in a vector format as rðtþ ¼h 0 xðtþ. We can get the mean value of the transmission process according to [17] r ¼ x X 1 t¼ 1 td 1 X h r ðtþ ¼ x h r ðtþ: It indicates that the mean value of rðtþ is equal to the mean of xðtþ times the area under the impulse response. As the sum of scheduling function h r ðtþ is one, the average transmission rate becomes r ¼ x. An intuitive explanation is that the transmitter should transmit all arrived packets within their deadlines because we assume no degradation in the transmission (sum of hðiþ is one). The objective of the energy optimization problem is to minimize IE½ðrðtÞ r Þ 2 Š¼IE½r 2 ðtþš 2 r ¼ IE½h0 xðtþh 0 xðtþš 2 x ¼ IE½h 0 xðtþx 0 ðtþhš 2 x ¼ h0 h 2 x : t¼0 ð4þ ð5þ ð6þ As we assume x is a constant, the problem is reduced to minimizing h 0 h. This is a typical optimization problem in signal processing with a linear constraint. It can be solved using Lagrange multipliers. The Lagrangian is Lðh;Þ¼h 0 h þ ðu h 1Þ; where u is a unitary vector with t d components and u h ¼ 1 is the vector form of the constraint (4). The gradient of L is 2h þ u 0 with a solution of h ¼ 1 u 0 : 2 To find the value of the Lagrange multiplier, the solution must satisfy the constraint (4). By imposing the constraint, we obtain a solution as in (5). It is the optimal time-invariant scheduling policy. tu This theorem reveals the impact of input autocorrelation structure on scheduling. The autocorrelation can be either measured online according to history packet arrivals or offline with a given input. In both cases, the autocorrelation can be computed by summing products of an input arrival with the same arrival delayed by a certain time lag [17]. A special case is independent packet arrivals in which the covariance matrix is diagonal and the solution in (5) becomes uniform. It means an equal amount of transmission is scheduled before the deadline. In this case, the scheduling policy according to Theorem 3.1 is the same as the optimal time invariant solution in [10]. We state this result in the following corollary: Corollary 3.1. If the input arrivals are independent over time, the optimal delay bounded packet scheduling function is h r ðtþ ¼1=t d, 0 t<t d. The transmission rate can be represented as rðtþ ¼ 1 t d X t t a ¼t t d þ1 xðt a Þ: 3.3 Example Solutions We next illustrate the solution by examining several traffic patterns with different autocorrelation functions (ACFs). The first is a multimedia trace of the popular Simpsons VBR video from [22]. It is a 20 minute clip consisting of 30,334 frames. The Hurst parameter is used to characterize the long range dependence (LRD) of the traffic. The degree of correlation is high with a large H. The Hurst parameter of the video clip is 0.84 and it has a strong degree of dependence and burstiness. The second is from a Fractional Gaussian Noise process [15] with a Hurst parameter 0.9. The last is a multimedia traffic model with its ACF based on a shifted exponential scene-length distribution [11], i.e., e jj, where ¼ 1=49 and is the time lag. Fig. 2 shows the ACFs of the traffic models with different lags. Their impacts on the optimal scheduling functions are shown in Fig. 3 with the delay constraint set to 10 time slots. A Poisson process, which is statistically independent, has been included for comparison. An interesting finding is that, as the degree of input autocorrelation increases (in the order of t d ), shown from Figs. 3a, 3b, 3c, and 3d, the ð7þ

4 ZHONG AND XU: ENERGY-EFFICIENT WIRELESS PACKET SCHEDULING WITH QUALITY OF SERVICE CONTROL 1161 Fig. 2. Autocorrelation of different input processes. convexity of the scheduling function also increases. This can be verified from the low-pass nature of the scheduling process. With increased degree of correlation, there are more low frequency components [1]. The low-pass filter needs to have a lower cutoff frequency to effectively smooth the input. As a result, the scheduling function for a Poisson process has the highest bandwidth while the lowest for the multimedia pattern. 4 QoS GUARANTEE A transmitter is characterized by two factors: maximum reliable transmission rate and queue size. Overload occurs when the required transmission rate exceeds the maximum rate. The energy-efficient scheduling policy represented by (3) tries to postpone packet transmission as late as possible under the delay constraint. As the scheduling is causal without knowledge of future packet arrivals, it is possible to get unexpected overload and packet drops by transmitting at a lower rate. Assume we have two arrivals from an independent process to a transmitter with a maximum transmission rate 100 packets per time slot. One arrives at time 0 with 100 packets; the other at time 1 with 400 packets. Both have a delay constraint 4. If we do not consider energy consumption and schedule transmission as fast as possible, the first arrival will be scheduled during time [0, 1) and the second during [1, 5), all with 100 packets per slot. This is a feasible schedule. If we use the energy efficient scheduler for an independent process, according to (7), the transmission for each arrival will be scheduled uniformly before its deadline. Thus, for the first arrival, 25 packets per time slot will be transmitted during [0, 4) and 100 packets during [1,5) for the second, as shown in Fig. 4b. We can see that the maximum required transmission rate is 125 packets per time slot. The schedule is no longer feasible in the system. To guarantee error-free packet transmission, we need to apply admission control policies to prevent system from being overloaded. We can either reject the packet arrivals that would otherwise overload the system or admit them but delay their transmissions. By the rejection policy in this example, we decline the second arrival and no deadline miss occurs. In contrast, by the admit-but-delay policy, the second arrival would not be transmitted before its deadline. As can be seen from Fig. 4c, the system cannot meet the deadline of the second arrival because it slows down the transmission of the first to save energy. Deadline miss cannot be totally avoided in energy efficient online packet scheduling without assumed knowledge of future packet arrivals. A similar observation can be made for a transmitter with a limited queue size. When we slow down packet transmission, more packets stay in the transmitter queue. As a result, there could be more queue overflows. Consider a transmitter with a queue size 300 packets. We show in Figs. 5b and 5c that both schedules under power control in Fig. 4 lead to queue overflow, while the power-oblivious schedule does not. The implication of queue overflow is that packets have to be dropped without being transmitted. The number of queue overflow is identical to the number of packet drops. Fig. 3. Unit impulse response of the transmission process with different ACFs. (a) Response with a Poisson process. (b) Response with a VBR video trace. (c) Response with an FGN process. (d) Response with a multimedia trace. Fig. 4. Impact of energy-efficient scheduling on delay guarantee. (a) Schedule without power control. (b) Schedule under power control with transmitter overload. (c) Schedule under power control with deadline miss.

5 1162 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 Fig. 5. Impact of energy-efficient scheduling on queue backlog. (a) Queue backlog without power control. (b) Queue backlog with queue overflow under schedule in Fig. 4b. (c) Queue backlog with queue overflow under schedule in Fig. 4c. To restrict deadline misses and packet drops in a controllable manner, we investigate the relationship between the maximum transmission rate and an overload probability and the relationship between the queue size and packet drop rate. Note that the four factors are interdependent. For example, an increase of transmission capacity leads to decreases of both overload probability and packet drop rate. The number of packet drops is reduced because more packets can be transmitted and fewer are buffered in the transmitter. It is very difficult to characterize the four factors in one expression. Our approach is to deal with them in two steps. Given a group of packet arrivals, we first determine the transmission capacity or the overload probability according to the analysis in Section 4.1. The basic idea is to obtain the transmission rate distribution that is required to transmit all packets before their deadlines. The transmission rate distribution is then used to provide a bound of the transmission capacity with an overload probability and vice versa. With derived transmission capacity or overload probability, we can then get input statistics to the queuing system and compute either the transmission queue size or the packet drop rate according to the results in Section Transmission Rate with Overload Probability We provide two bounds on the transmission capacity with a QoS constraint. The first bound makes no assumption about the arrival process while the second assumes independent input with known distribution. The former bound is loose as it holds for any input having the same input mean and variance and the latter is tight. We do not consider the possibility of buffer overflow in this subsection General Input Define the overload probability v r as probðrðtþ >r max Þ, where r max is the maximum reliable transmission rate. With known mean r and standard deviation r of the transmission rate rðtþ, we can estimate the probability distribution tail by Chebyshev s inequality as follows [7]: Theorem 4.1. The upper bound of the transmission rate that is required to guarantee a predefined overload probability v r for a general input is rffiffiffiffiffiffiffiffiffiffiffiffi 1 v r r max r þ r : ð8þ v r Proof. From Chebyshev s inequality, it is known that F fig a 1 IEðuðyÞÞ; where y is a random variable, I is an interval, FfIg is a distribution function, uðyþ >a>0 for all y in I. Substitute y with transmission rate rðtþ and define uðrðtþþ¼ðrðtþþxþ 2 with x>0. It can be verified that uðrðtþþ > ðr max þ xþ 2 > 0 for rðtþ >r max > 0. Therefore, Since we have probðrðtþ >r max Þ 1 ðr max þ xþ 2 IE½ðrðtÞþxÞ2 Š: IE½ðrðtÞþxÞ 2 Š¼ 2 r þ 2 r þ 2x r þ x 2 ; 1 probðrðtþ >r max Þ ðr max þ xþ 2 ðx2 þ 2x r þ 2 r þ 2 r Þ: It can be proved that the right side of the inequality takes the minimum value at x ¼ r þ 2 r =r max. Replace probðrðtþ >r max Þ with v r and compute the minimum value. We get 2 r v r 2 r þðr max r Þ 2 : ð9þ Solving the inequality for r max gets the rate bound. tu The theorem can be applied to determine the maximum transmission rate in the process of system design. Because of the high variability of input arrivals, the variance can be a dominant factor. For an online scheduler with a fixed maximum rate, we can provide a QoS control by applying (9). Practically no transmitter can transmit packets at a rate higher than its maximum rate. Admission control needs to be implemented when the system becomes overloaded. We can adopt two types of admission control policies:. Arrivals that could lead to overload are rejected. This makes sure that all admitted packets can be finished before their deadlines. The number of rejections would be the same as the number of overloading arrivals if there is no admission control. In this case, the rejection rate is equal to the overload probability.. Arrivals that could lead to overload are admitted, but not transmitted until the system has enough resource, similar to the task transformation technique in [24]. As a result, admitted packets may miss their deadlines. The number of misses is the same as the number of overloading arrivals if there is no admission control. In this case, the overload probability is equal to the deadline miss rate of the arrivals.

6 ZHONG AND XU: ENERGY-EFFICIENT WIRELESS PACKET SCHEDULING WITH QUALITY OF SERVICE CONTROL 1163 Under either policy, the overload probability can be used to bound the rejection rate or the deadline miss rate. We can also calculate the maximum transmission rate or overload probability from input statistics instead of from output. The mean and variance of the transmission process can be readily computed according to input in the LTI system. For a WSS input process, we recall from the proof of Theorem 3.1 that the mean value of the transmission rate equals the average arrival rate, r ¼ x. We only need to determine the variance of the transmission process. It can be computed according to [17] 2 r ¼ X1 X 1 t 1¼ 1 t 2¼ 1 ¼ Xt d 1 Xt d 1 t 1 ¼0 t 2 ¼0 R xx ðt 1 t 2 Þh r ðt 1 Þh r ðt 2 Þ 2 x R xx ðt 1 t 2 Þh r ðt 1 Þh r ðt 2 Þ 2 x ; ð10þ where R xx denotes the autocorrelation function of the input process. As h r ðtþ has a limited length determined by deadline t d, the computation cost is low Statistically Independent Input If the input process is WSS and statistically independent, we can determine the transmission rate with respect to input statistics as follows: Corollary 4.1. If the input process is WSS and statistically independent, with x and x as its mean and standard deviation, the relationship of deadline, maximum transmission rate, overload, and input statistics can be characterized by rffiffiffiffiffiffiffiffiffiffiffiffi 1 v r r max x þ x : ð11þ v r t d Proof. For independent input, the scheduling function should be set to a constant value h r ðtþ ¼1=t d during time 0 t<t d for minimizing transmission energy consumption. We can simplify the computation of the variance as 2 r ¼ 2 x Xt¼1 t¼ 1 h 2 r ðtþ ¼2 x Xt d 1 t¼0 h 2 r ðtþ ¼2 x t d : As r ¼ x, substituting r and r into Theorem 4.1 completes the proof. tu The bounds by Theorem 4.1 are loose as they are applicable to a general input. If more information about the input is available, we can get a tightened bound. In our prior work, we provided tightened bounds for a unimodal distribution [30]. In this work, we will show that, for a stationary input process, if the input arrivals at different time slots are independent and identically distributed (i.i.d.) with known distribution, we can get an exact analysis for the overload probability. The input process is assumed to be stationary because we require the probability distribution at a fixed time or position to be the same for all times or positions. We first determine the transmission rate distribution from input distribution and the delay constraints t d. The distribution is simply a convolution of t d input distribution scaled by 1=t d. Denote the probability density function (PDF) of input as fðxþ. The transmission rate distribution can be expressed as fðrþ ¼t t d d f t d ðt d xþ, where t d means t d -fold convolution of fðxþ with itself. As the convolution operation is computationally expensive, we next apply a transform method to simplify the computation of the output distribution. Denote x ð!þ as the characteristic function of the input PDF. We first get the characteristic function of the transmission rate distribution for a stationary input process, which is r ð!þ ¼IE½e j!rðtþ Š¼IE½e j!p t d 1 xðt iþhrðiþ Š ¼ Ytd 1 I¼0 ¼ Yt d 1 IE½e j!xðt iþhrðiþ Š¼ Ytd 1 x ðh r ðiþ!þ x ð!=t d Þ: ð12þ The PDF of rðtþ can then be found by taking the inverse Fourier transform of the product,! t Yd 1 fðrþ ¼F ð 1Þ x ð!=t d Þ : By considering the tail distribution of the transmission rate, we derive the relation between the maximum rate and an overload probability as Z 1 v r ¼ probðrðtþ >r max Þ¼ fðrþdr: r max We need to estimate PDF of incoming traffic. Generally, there are two ways to estimate distributions. One is a parametric method which assumes the input traffic belongs to a certain type of distribution and estimates parameters of the distribution. The other is a nonparametric method. A popular nonparametric method is kernel density estimation [23], which builds up a distribution by adding up distributions from sample points of the input. More details of estimation for specific distributions such as Normal, Gamma, and Pareto can be found in [13]. 4.2 Impact of Energy-Efficient Transmission on Packet Drop We have shown that the scheduling process can be modeled as a linear time-invariant (LTI) system. We next prove that with the scheduler as a LTI system, the queuing system can also be modeled as a linear system with a time-invariant transfer function. The model facilitates the analysis of the impact of energy-efficient transmission on queue backlog distribution and packet drop Queuing System Analysis Let sðtþ be the queue backlog length at the start of time t. The dynamics of the queue can be given by sðt þ 1Þ ¼sðtÞþxðt þ 1Þ rðtþ: ð13þ The rðtþ packets to be transmitted at time t are removed from the queue before the next xðt þ 1Þ packets arrive. Substituting rðtþ with (3) yields sðt þ 1Þ ¼sðtÞþxðt þ 1Þ xðtþh r ðtþ: ð14þ

7 1164 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 Fig. 6. Unit impulse response of the queuing system with different ACFs. (a) Response with a Poisson process. (b) Response with a VBR video. (c) Response with an FGN process. (d) Response with a multimedia trace. By taking z-transform of (14), we have zsðzþ ¼SðzÞþzXðzÞ XðzÞH r ðzþ; SðzÞ ¼ 1 H rðzþz 1 1 z 1 XðzÞ; ð15þ where SðzÞ, XðzÞ, and H r ðzþ are the z-transforms of sðtþ, xðtþ, and h r ðtþ, respectively. Let H s ðzþ ¼ð1 H r ðzþz 1 Þ=ð1 z 1 Þ. It follows that SðzÞ ¼H s ðzþxðzþ. Given the packet scheduling process modeled as a LTI system with a transfer function h r ðtþ, the queuing system can also be represented as a LTI system with a transfer function h s ðtþ with H s ðzþ as its z-transform. We refer to this function as a queuing function. With an input process, we can determine h r ðtþ according to (5). To determine h s ðtþ, a straightforward way is to first get the z-transform of h r ðtþ and then compute H s ðzþ according to (15). Once H s ðzþ is determined, we can get h s ðtþ by taking the inverse z-transform of H s ðzþ. Alternatively, we can calculate the unit impulse response of the queuing system directly from h r ðtþ, which is much easier. Theorem 4.2. The queuing system can be modeled as an LTI system given that the scheduling process is an LTI system with a transfer function h r ðtþ. The transfer function of the queuing system, h s ðtþ, can be calculated by h s ðtþ ¼1 Xt 1 h r ðiþ: ð16þ Proof. According to (15), we know that the queuing system can be modeled as an LTI system with a transfer function H s ðzþ (or h s ðtþ equivalently). We next prove the expression of h s ðtþ by induction. When t ¼ 0, there is an input arrival with size 1 and the queue is empty. The queue backlog is simply the current arrival. We define h s ð0þ as 1. Starting from t ¼ 1, we derive the queue backlog according to (13) as h s ð1þ ¼h s ð0þþxð1þ rð0þ ¼1 h r ð0þ: Therefore, (16) holds for t ¼ 1, which establishes a basis for the inductive argument. Suppose that the formula holds for t 1: h s ðt 1Þ ¼1 Xt 2 h r ðiþ: Substituting it into (13) yields h s ðtþ ¼h s ðt 1ÞþxðtÞ rðt 1Þ! ¼ 1 Xt 2 h r ðiþ þ 0 h r ðt 1Þ ¼1 Xt 1 h r ðiþ: By induction, the result holds for all t 0. Note that, as t t d, since the arrival has been transmitted and there is no new input, the queue backlog h s ðtþ ¼0. We can verify this from (16) by noting P t d 1 h rðiþ ¼1. tu We have shown that both the queuing system and the transmission process can be modeled as LTI systems. The major difference is that they have different transfer functions, h s ðþ and h r ðþ. Fig. 6 plots the queuing functions with respect to different input described in Section 3.3. With the increase of input correlation, the queuing functions turn flat. This is consistent with existing findings that the presence of input correlation makes it less effective to smooth input using a queue. As a result, a higher cutoff frequency is needed for input with higher correlation. For example, in Fig. 6d, the queuing function for the multimedia pattern has the highest bandwidth. In a similar way to the transmission rate and overload analysis, we derive the relation between queue size and an overflow probability. Define the probability as v s ¼ probðsðtþ >s max Þ, where s max is the queue size. We use the overflow probability to bound packet drop rate. Let s and s denote the mean and standard deviation of the queue backlog. It is not difficult to prove similar results as in Theorem 4.1 and (10) by changing r max, v r, r, r, h r ðþ to s max, v s, s, s, h s ðþ, respectively. Details are omitted for brevity Statistically Independent Input When input is WSS and statistically independent, we can characterize the capacity-qos relation with the following corollary: Corollary 4.2. If the input process is WSS and statistically independent, an upper bound of the queue size with respect to delay constraint and an overflow probability under the delay bounded packet scheduling is rffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 v s ðt d þ 1Þð2t d þ 1Þ s max x þ t d þ 1 x : ð17þ v s 6t d 2

8 ZHONG AND XU: ENERGY-EFFICIENT WIRELESS PACKET SCHEDULING WITH QUALITY OF SERVICE CONTROL 1165 Proof. We derive the queue backlog mean as! X t d 1 t Xd 1 s ¼ x h s ðiþ ¼ x 1 Xi 1 h r ðjþ ¼ x X td 1! 1 Xi 1 1 t j¼0 d j¼0 ¼ t d þ 1 x : 2 ð18þ Similarly, the variance of the backlog is! 2 s ¼ VarðsðtÞÞ ¼ Var Xtd 1 ¼ Xtd 1 ¼ 2 x Varðxðt iþþh 2 s ðiþ ¼2 x Xt d 1! 2 1 Xi 1 h r ðjþ ¼ 2 j¼0 ¼ ðt d þ 1Þð2t d þ 1Þ 6t d 2 x : xðt iþh s ðiþ Xt d 1 Xt d 1 x h 2 s ðiþ 1 Xi 1 1 t j¼0 d! 2 ð19þ By adapting Theorem 4.1 to the queuing system, we rewrite (8) as rffiffiffiffiffiffiffiffiffiffiffiffi 1 v s s max s þ s : ð20þ v s Combining (18) and (19) with (20) gets an estimate of queue size in (17). tu If the input process is stationary and i.i.d. with known distribution, we can get a tight queue backlog bound by first deriving the PDF of the backlog. The computation can be simplified by first deriving its characteristic function, similar to the transmission rate analysis. The characteristic function of the queue backlog distribution is!! s ð!þ ¼ Ytd 1 x ðh s ðiþ!þ ¼ Ytd 1 1 Xi 1 h r ðjþ! ¼ Ytd 1 x t d i! : t d x j¼0 ð21þ We finally note that the choice of admission control policies has impact on the input to the queuing system. If we reject all packets that may lead to overload, the queuing system may have less number of input arrivals; if we admit packets that cannot be transmitted before their deadlines and transmitted them in a best-effort manner, the queuing system may have a larger number of packets. However, the analytical results in this section still hold with an altered input process to the queuing system. 5 PERFORMANCE EVALUATION We conducted simulations to verify the analytical results. The simulations were designed in three aspects: 1) Investigate the effectiveness of the proposed scheduling policy for energy saving, 2) demonstrate the use of the capacity bounds for QoS control both in offline configuration and online scheduling, and 3) how the effectiveness of the Fig. 7. Energy consumption with different delay constraints. capacity-qos relationships in dealing with input variation. We assumed a packet length of 8 KBits and a channel capacity of 8 bits/transmission with a channel bandwidth B ¼ transmissions/s. Under this setting, the minimum transmission duration for a packet is 2 ms, which is the time granularity of the transmitter. The power function is according to 1), PðrðtÞÞ ¼ 2 2rðtÞ=5105 1, with the noise power set to Effectiveness in Energy Savings The first experiment was conducted with an input of video traces from [22]. An important characteristic of the traces is their time-correlation. To show the performance of the schedulers under traffic with different degrees of autocorrelation, we chose two VBR video traces, JurassicPark I and Simpsons, with Hurst parameters 0.92 and 0.84, respectively. A baseline packet scheduling policy [10] is implemented for comparison. It is a special case of our proposed policy for statistically independent input process with a uniform scheduling function. To compare energy consumption using the proposed scheduling policy with the uniform allocation, we simulated transmission of the video traces. Fig. 7 plots the energy consumptions with different settings of delay constraints with the baseline policy referred as Uniform. The energy savings by both schedulers increases with larger delay constraints. This is because the transmission rate is reduced with relaxed delay constraints. Due to the convexity of the power function, more energy is saved. It is clear that the proposed scheduler determined by input correlation consistently outperforms the uniform allocation. The improvement is up to 15 percent for JurassicPark and 8 percent for Simpsons. The results of JurassicPark show more benefits by consideration of input correlation. This is expected because high input correlation makes a uniform allocation less effective in energy savings. 5.2 Capacity Configuration We next investigated the effectiveness of the bounds in offline capacity configuration subject to a QoS constraint for a general input (8), (17), the tight bounds when input arrivals are statistically independent with a known distribution (12), (21). We generated the number of packets to be transmitted at each time slot by a Gaussian distribution nð20; 4Þ. The length of each time slot is assumed to be 40 ms. We first set the delay constraint to 10 time slots and plot transmission rates with different overload probabilities in Fig. 8. Expectedly, the capacity requirement increases with a

9 1166 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 Fig. 8. Maximum transmission rate with different overload probabilities. Fig. 11. Empirical CDFs of queue size from theoretic analysis and simulation. Fig. 9. Empirical CDFs of transmission rate from theoretic analysis and simulation. Fig. 12. Maximum transmission rate with different delay constraints. Fig. 10. Queue size with different overflow probabilities. smaller chance of failure. The bounds based on general input are loose as they hold for all input distributions with the same mean and variance. The bounds by (12) are tight with known input distribution. We also estimated the CDF of transmission rate by simulation using a histogram technique. We compare the obtained CDF with that from the analytical results (12) in Fig. 9; the resultant capacity bounds from the CDFs according to an overload rate 1 percent are shown in Fig. 8. We can observe that both the bounds and the CDF from the analysis agree quite well with the simulation results. Similar observations can be made with the relationship between queue size and an overflow probability in Fig. 10 and the CDFs of queue size in Fig. 11. The bounds of transmission rate with different delay constraints are shown in Fig. 12 with overload probabilities of 1 percent and 10 percent. The transmission rate requirement is reduced greatly with increased delay when the delay is smaller than 10 time units. With larger delay, the transmission bound reduces at a slower speed. Note that the transmission rate requirement is an upper bound of the transmission rate, which is determined by the mean and Fig. 13. Queue size with different delay constraints. variance of the rate distribution. A theoretical lower bound of the transmission rate is the average arrival rate (4 Mbps), below which, the system is not stable. If the delay is large enough, variance of the output rate approaches zero and the upper bound converges to the lower bound. Fig. 12 also reveals that the upper bounds with a large overload probability of 10 percent tend to flatten out faster than those with a small probability of 1 percent. This is because a smaller overload probability has a high capacity demand and is more dependent on the transmission rate variance. As the delay increases, packets stay in the queue for a longer time and a larger queue size is required, as can be observed from Fig. 13. It is interesting to find that the bounds can be well approximated by a linear relationship with delay constraint. A close examination of (17) with different parameters indicates that the higher order coefficients of t d are much smaller than the linear coefficient; for example, the coefficient of t 2 d is only 0.16 percent of t d in most cases. The linear approximation can be used to simplify queue size configuration with different delay constraints.

10 ZHONG AND XU: ENERGY-EFFICIENT WIRELESS PACKET SCHEDULING WITH QUALITY OF SERVICE CONTROL 1167 Fig. 14. Normal plots of input, transmission rate, and queue backlog. We point out that when the marginal distribution of the input process is Gaussian, the transmission rate rðtþ and queue backlog sðtþ also follow a Gaussian distribution. This can be verified by the normal plots of input, output rate, and queue backlog in Fig. 14. Note that the normal plot for data from a Gaussian distribution is a straight line. The figure shows that both rðtþ and sðtþ are Gaussian with a Gaussian arrival. This greatly simplifies the capacity-qos analysis because the Gaussian distribution is totally characterized by its mean and variance, which can be readily computed according to Section 4. We also observe that the slope of the transmission rate plot is steeper than that of the input. This is consistent with the low pass nature of the scheduler, which smooths the fluctuating traffic with a smaller output variance and leads to better schedulability and QoS assurance. 5.3 Qos Control In the next experiments, we show the effectiveness of the capacity-qos relationships in providing statistical QoS control during online packet scheduling. We used the same set of parameters as the last experiment, with number of input packets per time slot following a Gaussian distribution nð20; 4Þ, each with a delay constraint of 10. We limited the chance of failure to 1 percent for both overload and overflow. As we do not allow packet transmission at a speed higher than the maximum reliable rate and the queue backlog cannot exceed the queue size, we enhanced the scheduler in handling overload and overflow. In case of transmitter overload, the system transmits under the maximum rate and the unfinished packets are scheduled for transmission in a best-effort mode. As a result, packet transmission may not finish before its deadline. We characterized the QoS degradation with deadline miss rate and response time in the experiments. In case of queue overflow, we applied admission control to drop the packets that cannot be put into the queue. We used drop rate to characterize the failure due to queue overflow. We first experimented with the loose capacity bounds, i.e., 7 Mbps and 188 KBytes for maximum transmission rate and queue size, as can be observed from Figs. 8 and 10. We found no deadline misses or packet drops. This shows the conservativeness of the estimate because of the target 1 percent chance of failure. To show the effectiveness of the tight capacity bounds, we experimented with the tight bound for the transmission rate, 4.58 Mbps. Both loose (188 KBytes) and tight (128 KBytes) queue size bounds are experimented with. The simulation results are shown in Table 1. We can observe that there are deadline misses under the tight transmission rate even when the queue size is large enough, under which no packet drop occurs. The observed 0.88 percent miss rate is close to the 1 percent simulation setting. When we further set a tight queue size, both deadline misses and packet drops occur. The occasional deadline misses cause slightly larger response times than the target 10 units. However, the degradation is not much since more than 99 percent of the requests meet their deadlines, the average response time is close to 10, and the variance of the response time is small. The mean and variance of the transmission rate and queue size are smaller with the tight queue bound due to the 0.71 percent packet drop. 5.4 Qos Control under Traffic Variation We have shown that QoS control can be provided by capacity configuration on the transmitter according to input statistics. If the input traffic is not a WSS process, out analysis cannot be applied. In such a case, the service might degrade. Interesting questions are, when there are more arrivals than expected, what capacity is required to keep the same level of QoS or how much degradation there will be with the same capacity. The answer to the first question shows how capacity configuration can be used to keep the same level of QoS. Following the previous experiment with more variable input arrivals, we generate input with a Gaussian distribution nð20; 6Þ. To keep the original deadline of 10 time units and a 1 percent overload probability, we need a higher reliable transmission rate. According to Corollary 4.1, we have 2 2 ðr max1 1 Þ ðr max2 2 Þ 2 ¼ 1: As a result, the new transmission rate is r max2 ¼ 4:87 Mbps. Similarly, we determined the queue size according to (17) as 227 KBytes. We ran simulations based on the new predicted capacities. The experimental results in the second row in Table 2 show that, with the adjusted rate and queue size, 1.1 percent of the packets miss their deadlines close to the target QoS level. Alternatively, we can adapt the QoS target without changing the transmission rate. The new delay constraint t 2, TABLE 1 Scheduling with a Tight Transmission Rate for nð20; 4Þ ðr max ¼ 4:58 Mbps, t d ¼ 10Þ

11 1168 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 TABLE 2 Scheduling with Increased Maximum Transmission Rate or Delay Constraint for nð20; 6Þ based on the target 1 percent overload probability for nð20; 6Þ, is t 2 ¼ t 1 ¼ : 42 The increase of delay constraint requires a larger queue size to hold more delayed packets. As a result, the queue size has to be increased to 410 KBytes according to (17). In practice, if the queue size is the bottleneck, we can change the target delay constraint online to adapt with the variation of input traffic. We ran simulations with a new deadline of 23 time units. The capacity was kept to 4:58 Mbps, an upper bound as shown in Fig. 8. The results are in the third row of Table 2. We can observe that, in response to the change of incoming traffic, about 99.2 percent of the packets meet their adjusted deadline of 23. This again demonstrates the prediction accuracy of the analysis results. 6 RELATED WORK Different approaches have been proposed to investigate the trade-off between power and delay. Average queuing delay was considered in [5], [2], [19]. Collins and Cruz proposed an optimal transmission scheme in a fading channel with an average delay constraint and a peak transmitter power [5]. They used a simplistic channel model with a two-state Markov chain and assumed that energy expenditure is linear with transmitted data. Berry and Gallager considered the energy minimization problem with an average queuing delay in a block-fading channel [2]. The energy minimization was turned into a convex optimization problem and dynamic programming was used to find the optimal solution. Rajan et al. studied the energy gain due to source burstiness with mean queuing delay constraints [19]. An optimal scheduler obtained by dynamic programming and a near-optimal approximation was proposed for energy savings over both Gaussian and fading channels. They assumed independent and identically distributed (i.i.d.) packet arrivals at each time slot. They pointed out that i.i.d. source models rarely represent real sources, which usually have a strong time-correlation. However, they focused on the applicability of energy-efficient packet scheduling in a real system and did not study the impact of the correlation. There are also other similar metrics with average delay. For example, Schurgers et al. studied the impact of channel state under an average data rate constraint [21]. A different delay constraint was applied in [25], [9], [8], [29], in which a single deadline was put to all packets. The indirect bound on packet delay requires that all packets arrive before T to be transmitted no later than T. Offline optimal and online near-optimal algorithms were proposed for a single transmitter-receiver pair by Uysal-Biyikoglu et al. [25]. The authors showed that their proposed scheduler is more energy-efficient than a deterministic constant service policy that results in the same average delay. In their later extension to multiple users, they applied the same delay constraint [9]. Fu et al. proposed a packet transmission policy to send an amount of data within a fixed time period [8]. Their focus was on the impact of fading channel on throughput and energy optimization. Zhang and Chanson targeted maximum system throughput and value in a Gaussian channel under energy and time constraints [29]. Nuggehalli et al. [16] utilized both average delay and deadline constraints, considering the effect of energy recovery during idle periods. They first adapted the scheduling policy in [25] with a battery recovery model and reported more than 50 percent energy savings. They then applied average delay and derived the optimal solution with and without battery recovery. They showed that the battery-aware policy could save substantial energy. A more practical and widely used constraint for delaysensitive applications is an explicit deadline for each packet [21], [19], [10], [14], [28], [3]. Schurgers et al. analyzed energy-efficient real-time packet scheduling in a timeinvariant channel with a detailed uncoded MQAM modulation [21]. They used a sufficient schedulability test to assure no deadline misses. As the condition is only sufficient, not necessary, it is possible to reject more tasks than necessary. They noted that it is too computationally intensive to compute an energy optimized schedule and proposed a heuristic algorithm. However, it is not clear how efficient their heuristics is and, most importantly, the usage of their approach is limited only to a periodic task model with constant packet interarrival times. Rajan et al. [19] used an absolute deadline to each packet for an i.i.d input. They formalized the optimization problem and solved using a value iteration algorithm. However, the computational complexity of their algorithm grows exponentially with the delay constraint and the complexity is prohibitive with delay larger than three units. This limits its use in an environment with limited power. In contrast, the proposed policy in this paper has a low time complexity. Khojastepour and Sabharwal considered a strict maximum delay constraint for each packet [10]. They established the connection between maximum delay scheduling and a linear filter for an i.i.d. input. Two optimal scheduling approaches were proposed. One is a time-variant policy which makes scheduling decisions according to each new packet arrival and uncompleted arrivals in the queue backlog. Each time a packet arrives, all arrivals in the backlog must be iterated in order for the algorithm to derive

12 ZHONG AND XU: ENERGY-EFFICIENT WIRELESS PACKET SCHEDULING WITH QUALITY OF SERVICE CONTROL 1169 a new transmission rate for the new arrival and updated rates for those in the backlog. Scalability might be a concern when there are a large number of packet arrivals. In contrast, the time-invariant policy for time independent input in [10] has a constant running time in determining a transmission rate for a packet arrival. In addition, the timevariant policy works more aggressively by setting the packet transmission at a lower rate. It may lead to more deadline misses. Because of its time-variant nature, any analysis of the deadline miss rate according to input is hard, even for time independent input. We therefore focus on the time-invariant policy for time correlated input and provide a QoS control to bound the number of deadline misses or rejections. Mangharam et al. [14] provided QoS support for multiple users with bursty MPEG-4 video transmission over a time-varying channel. They also used deadline miss rate as a QoS requirement and proposed an online greedy algorithm to satisfy the constraint. However, they considered a different scenario in which there is a centralized Access Point (AP). The AP assigns a transmission rate to each node such that the QoS constraints for multiple users are met. In our work, we restrict our focus on communications between a pair of senders and receivers. Zafer and Modiano presented a generalization of energyefficient packet scheduling by a calculus approach [28]. They proposed an offline generalization of the energy minimization problem and an online scheduler for a Poisson arrival. The online solution can be interpreted as the time-invariant scheduler in [10] plus anticipation of future arrivals. The scheduler can achieve less energy expenditure with Poisson input. An offline algorithm was also proposed by Chen and Mitra for a Gaussian channel and proved optimal with individual packet delay [3]. Their algorithm is a generalization of [25] by changing the group deadline to individual deadline. The authors showed that their algorithm can be fully characterized if the input is a Poisson process; for example, the optimal transmission durations of all the packets can be derived. We point out that the knowledge of all packet arrivals is not usually available offline and an online algorithm is more appropriate in dealing with the dynamic nature of input arrivals. In addition, the assumption that the input is a Poisson process is limited as the input is usually time correlated and more bursty. Finally, our work can also be distinguished by providing a statistical QoS guarantee. We studied the impact of slow transmission rate on both deadline misses and queue overflows and provided statistical control of service degradation. The provisioning of a statistical QoS control has been widely investigated in the real-time community; see [6] and [27] for examples of energy-oblivious and energy-aware analysis, respectively. Their approaches are to measure the uncertainty of execution time demand and adjust the scheduling policy based on the distribution information. They are not applicable to packet scheduling because packet size can be determined once a packet is admitted, while, in the CPU counterpart, the exact computation time of a task is not available even after task release. In our approach, instead of statistical analysis of input time demand, we analyze the tail distribution of output transmission rate and queue backlog. It is along the line of recent advances in statistical real-time guarantee for CPU using the Earliest Deadline First scheduling [31]. We finally point out that there exists work on the energyaware selection of appropriate computation speed and onchip buffer sizes in System-on-Chip (SoC) design. One recent example is due to Liu et al. [12], in which a calculus approach was used to determine CPU speed and buffer size from a given input. Our work is complementary because it supports transmitter rate and queue size configuration for a wireless interface in SoC design. The proposed approach is general in the sense that it can be applied to a group of inputs having the same statistical characteristics rather than to only a specific input. 7 CONCLUSION We consider minimal energy transmission over AWGN channels with delay constraints for each packet. In light of the fact that the input process for a wireless transmitter may be time-correlated, we consider packet scheduling for a general input process without a priori knowledge of input distribution. An energy-efficient scheduling policy has been proposed to take into account input autocorrelation, in which independent process is only a special case. As a slow transmission may result in unexpected overload and packet drops, we provide a statistical QoS control by investigating the relationship between transmitter capacity and a QoS constraint. We reveal the inherent relationships between the maximum reliable transmission rate and an overload probability, between the queue size and a packet drop rate. The relationships hold for all input distributions with the same first and second order moments. A tight bound is derived when the input process is time independent with a known distribution. The relationship can be used offline for capacity configuration during system design with a target QoS constraint and online by giving a service degradation bound with a fixed configuration. Simulation results show an improvement up to 15 percent over a recent policy for independent input. We demonstrate the effectiveness of the bounds in offline capacity configuration, online QoS control, and adaptation to input variation. We note that the channel model considered in this paper is an AWGN channel. We will investigate the impact of more realistic wireless fading channels and multiuser environments in our future work. ACKNOWLEDGMENTS This research was supported in part by US National Science Foundation grants ACI , CCF , and DMS and NASA grant 03-OBPR The authors thank the reviewers for their valuable suggestions. REFERENCES [1] J. Beran, R. Sherman, M. Taqqu, and W. Willinger, Long-Range Dependence in Variable Bit-Rate Video Traffic, IEEE/ACM Trans. Comm., vol. 43, nos , pp , Feb./Mar./Apr

13 1170 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 10, OCTOBER 2007 [2] R.A. Berry and R.G. Gallager, Communication over Fading Channels with Delay Constraints, IEEE Trans. Information Theory, vol. 48, no. 5, pp , [3] W. Chen and U. Mitra, Energy Efficient Scheduling with Individual Packet Delay Constraints, Proc. INFOCOM, [4] K.-H. Cho and H. Samueli, A 8.75-Mbaud Single-Chip Digital Qam Modulator with Frequency-Agility and Beamforming Diversity, Proc. IEEE Custom Integrated Circuits Conf., pp , [5] B. Collins and R.L. Cruz, Transmission Policies for Time Varying Channels with Average Delay Constraints, Proc. Allerton Conf. Comm., Control, and Computing, [6] J.L. Díaz, D.F. García, K. Kim, C.-G. Lee, L.L. Bello, J.M. López, S.L. Min, and O. Mirabella, Stochastic Analysis of Periodic Real- Time Systems, Proc. IEEE Real-Time Systems Symp., [7] W. Feller, An Introduction to Probability Theory and Its Applications, vol. 2. John Wiley and Sons, Inc., [8] A. Fu, E. Modiano, and J.N. Tsitsiklis, Optimal Energy Allocation for Delay-Constrained Data Transmission over a Time-Varying Channel, Proc. INFOCOM, [9] A.E. Gamal, C. Nair, B. Prabhakar, E. Uysal-Biyikoglu, and S. Zahedi, Energy-Efficient Scheduling of Packet Transmissions over Wireless Networks, Proc. INFOCOM, [10] M.A. Khojastepour and A. Sabharwal, Delay-Constrained Scheduling: Power Efficiency, Filter Design, and Bounds, Proc. INFOCOM, Mar [11] M. Krunz and A.M. Ramasamy, The Correlation Structure for a Class of Scene/Based Video Models and Its Impact on the Dimensioning of Video Buffers, IEEE Trans. Multimedia, vol. 2, no. 1, pp , [12] Y. Liu, S. Chakraborty, and W.T. Ooi, Approximate VCCS: A New Characterization of Multimedia Workloads for System-Level MPSOC Design, Proc. Design Automation Conf., pp , [13] J.R. Lorch and A.J. Smith, Improving Dynamic Voltage Scaling Algorithms with PACE, Proc. ACM SIGMETRICS Conf., pp , [14] R. Mangharam, S. Pollin, B. Bougard, R. Rajkumar, F. Catthoor, L.V. der Perre, and I. Moeman, Optimal Fixed and Scalable Energy Management for Wireless Networks, Proc. INFOCOM, [15] I. Norros, A Storage Model with Self-Similar Input, Queueing Systems, vol. 16, pp , [16] P. Nuggehalli, V. Srinivasan, and R.R. Rao, Delay Constrained Energy Efficient Transmission Strategies for Wireless Devices, Proc. INFOCOM, [17] P. Peebles Jr., Probability, Random Variables, and Random Signal Principles. McGraw Hill, [18] V. Raghunathan, C.L. Pereira, M.B. Srivastava, and R.K. Gupta, Energy Aware Wireless Systems with Adaptive Power-Fidelity Tradeoffs, IEEE Trans. Very Large Scale Integration (VLSI) System, vol. 13, no. 2, pp , [19] D. Rajan, A. Sabharwal, and B. Aazhang, Delay-Bounded Packet Scheduling of Bursty Traffic over Wireless Channels, IEEE Trans. Information Theory, vol. 50, no. 1, pp , [20] C. Schurgers, O. Aberthorne, and M.B. Srivastava, Modulation Scaling for Energy Aware Communication Systems, Proc. Int l Symp. Low Power Electronics and Design, pp , [21] C. Schurgers, V. Raghunathan, and M.B. Srivastava, Power Management for Energy-Aware Communication Systems, ACM Trans. Embedded Computing Systems, vol. 2, no. 3, pp , [22] P. Seeling, M. Reisslein, and B. Kulapala, Network Performance Evaluation Using Frame Size and Quality Traces of Single-Layer and Two-Layer Video: A Tutorial, IEEE Comm. Surveys and Tutorials, vol. 6, no. 2, pp , [23] B.W. Silverman, Density Estimation for Statistics and Data Analysis. Chapman and Hall, [24] T.-S. Tia, Z. Deng, M. Shankar, M. Storch, J. Sun, L.-C. Wu, and J.W.-S. Liu, Probabilistic Performance Guarantee for Real-Time Tasks with Varying Computation Times, Proc. IEEE Real Time Technology and Applications Symp., pp , [25] E. Uysal-Biyikoglu, B. Prabhakar, and A.E. Gamal, Energy- Eficient Packet Transmission over a Wireless Link, IEEE/ACM Trans. Networking, vol. 10, no. 4, pp , [26] W. Willinger, M.S. Taqqu, R. Sherman, and D.V. Wilson, Self- Similarity through High-Variability: Statistical Analysis of Ethernet LAN Traffic at the Source Level, IEEE/ACM Trans. Networking, vol. 5, no. 1, pp , [27] W. Yuan and K. Nahrstedt, Energy-Efficient Soft Real-Time CPU Scheduling for Mobile Multimedia Systems, Proc. 19th ACM Symp. Operating Systems Principles, [28] M. Zafer and E. Modiano, A Calculus Approach to Minimum Energy Transmission Policies with Quality of Service Guarantees, Proc. INFOCOM, [29] F. Zhang and S.T. Chanson, Throughput and Value Maximization in Wireless Packet Scheduling under Energy and Time Constraints, Proc. IEEE Real-Time Systems Symp., pp , [30] X. Zhong and C.-Z. Xu, Delay-Constrained Energy-Efficient Wireless Packet Scheduling with QoS Guarantees, Proc. IEEE Global Telecomm. Conf., [31] X. Zhong and C.-Z. Xu, Energy-Aware Modeling and Scheduling for Dynamic Voltage Scaling with Statistical Real-Time Guarantee, IEEE Trans. Computers, vol. 56, no. 3, pp , Mar Xiliang Zhong received the BS degree in radio engineering from Southeast University, China, in 1997 and the MS degree in electrical engineering from the Beijing University of Posts and Telecommunications, China, in He worked as a software engineer in the Zhongxing Telecom Corporation from 2000 to He is a PhD student in the Department of Electrical and Computer Engineering, Wayne State University, Detroit, Michigan. His current research interests include power management in embedded systems, mobile computing, and resource management in distributed systems. He is a student member of the IEEE. Cheng-Zhong Xu received the BS and MS degrees in computer science from Nanjing University in 1986 and 1989, respectively, and the PhD degree in computer science from the University of Hong Kong in He is an associate professor in the Department of Electrical and Computer Engineering at Wayne State University. His research interests lie in distributed and parallel systems, particularly in scalable and secure Internet services, adaptive and highly reliable networked computer systems, and resource management in cluster and grid computing. He has published more than 100 peerreviewed articles in journals and conference proceedings in these areas. He is the author of the book Scalable and Secure Internet Services and Architecture (Chapman & Hall/CRC Press, 2005) and a coauthor of the book Load Balancing in Parallel Computers: Theory and Practice (Kluwer Academic/Springer-Verlag, 1997). He serves on the editorial boards of the Journal of Parallel and Distributed Computing, the Journal of Parallel, Emergent, and Distributed Systems, the Journal of High Performance Computing and Networking, and the Journal of Computers and Applications. He was the founding program cochair of the International Workshop on Security in Systems and Networks (SSN), the general cochair of the IFIP 2006 International Conference on Embedded and Ubiquitous Computing (EUC 06), and a member of the program committees of numerous conferences. His research was supported in part by the US National Science Foundation and NASA. He is a recipient of the Faculty Research Award of Wayne State University in 2000, the President s Award for Excellence in Teaching in 2002, and the Career Development Chair Award in He is a senior member of the IEEE.. For more information on this or any other computing topic, please visit our Digital Library at

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 50, NO. 12, DECEMBER Elif Uysal-Biyikoglu, Member, IEEE, and Abbas El Gamal, Fellow, IEEE

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 50, NO. 12, DECEMBER Elif Uysal-Biyikoglu, Member, IEEE, and Abbas El Gamal, Fellow, IEEE IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 50, NO. 12, DECEMBER 2004 3081 On Adaptive Transmission for Energy Efficiency in Wireless Data Networks Elif Uysal-Biyikoglu, Member, IEEE, Abbas El Gamal,

More information

Maximizing the Number of Users in an Interactive Video-on-Demand System

Maximizing the Number of Users in an Interactive Video-on-Demand System IEEE TRANSACTIONS ON BROADCASTING, VOL. 48, NO. 4, DECEMBER 2002 281 Maximizing the Number of Users in an Interactive Video-on-Demand System Spiridon Bakiras, Member, IEEE and Victor O. K. Li, Fellow,

More information

Delay-minimal Transmission for Energy Constrained Wireless Communications

Delay-minimal Transmission for Energy Constrained Wireless Communications Delay-minimal Transmission for Energy Constrained Wireless Communications Jing Yang Sennur Ulukus Department of Electrical and Computer Engineering University of Maryland, College Park, M0742 yangjing@umd.edu

More information

CHAPTER 5 PROPAGATION DELAY

CHAPTER 5 PROPAGATION DELAY 98 CHAPTER 5 PROPAGATION DELAY Underwater wireless sensor networks deployed of sensor nodes with sensing, forwarding and processing abilities that operate in underwater. In this environment brought challenges,

More information

Introduction to Real-Time Communications. Real-Time and Embedded Systems (M) Lecture 15

Introduction to Real-Time Communications. Real-Time and Embedded Systems (M) Lecture 15 Introduction to Real-Time Communications Real-Time and Embedded Systems (M) Lecture 15 Lecture Outline Modelling real-time communications Traffic and network models Properties of networks Throughput, delay

More information

TRAFFIC from video services, especially videoconference

TRAFFIC from video services, especially videoconference IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 8, NO. 9, SEPTEMBER 2009 1153 Dynamic versus Static Traffic Policing: A New Approach for Videoconference Traffic over Wireless Cellular Networks Polychronis

More information

Loss Performance Analysis for Heterogeneous

Loss Performance Analysis for Heterogeneous IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 10, NO. 1, FEBRUARY 2002 125 Loss Performance Analysis for Heterogeneous ON OFF Sources With Application to Connection Admission Control Guoqiang Mao, Student

More information

Performance of UMTS Radio Link Control

Performance of UMTS Radio Link Control Performance of UMTS Radio Link Control Qinqing Zhang, Hsuan-Jung Su Bell Laboratories, Lucent Technologies Holmdel, NJ 77 Abstract- The Radio Link Control (RLC) protocol in Universal Mobile Telecommunication

More information

END-TO-END estimation of the spare capacity along a network

END-TO-END estimation of the spare capacity along a network 130 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 16, NO. 1, FEBRUARY 2008 A Stochastic Foundation of Available Bandwidth Estimation: Multi-Hop Analysis Xiliang Liu, Kaliappa Ravindran, and Dmitri Loguinov,

More information

Energy-efficient Transmission over a Wireless Link via Lazy Packet Scheduling

Energy-efficient Transmission over a Wireless Link via Lazy Packet Scheduling Energy-efficient Transmission over a Wireless Link via Lazy Packet Scheduling Balaji Prabhakar Elif Uysal Biyikoglu and Abbas El Gamal Information Systems Laboratory Stanford University Stanford CA 35

More information

Perspectives on Network Calculus No Free Lunch but Still Good Value

Perspectives on Network Calculus No Free Lunch but Still Good Value ACM Sigcomm 2012 Perspectives on Network Calculus No Free Lunch but Still Good Value Florin Ciucu T-Labs / TU Berlin Jens Schmitt TU Kaiserslautern Outline Network Calculus (NC): A Theory for System Performance

More information

Medium Access Control Protocols With Memory Jaeok Park, Member, IEEE, and Mihaela van der Schaar, Fellow, IEEE

Medium Access Control Protocols With Memory Jaeok Park, Member, IEEE, and Mihaela van der Schaar, Fellow, IEEE IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 18, NO. 6, DECEMBER 2010 1921 Medium Access Control Protocols With Memory Jaeok Park, Member, IEEE, and Mihaela van der Schaar, Fellow, IEEE Abstract Many existing

More information

Buffer Management for Self-Similar Network Traffic

Buffer Management for Self-Similar Network Traffic Buffer Management for Self-Similar Network Traffic Faranz Amin Electrical Engineering and computer science Department Yazd University Yazd, Iran farnaz.amin@stu.yazd.ac.ir Kiarash Mizanian Electrical Engineering

More information

Reduction of Periodic Broadcast Resource Requirements with Proxy Caching

Reduction of Periodic Broadcast Resource Requirements with Proxy Caching Reduction of Periodic Broadcast Resource Requirements with Proxy Caching Ewa Kusmierek and David H.C. Du Digital Technology Center and Department of Computer Science and Engineering University of Minnesota

More information

INTERNET traffic often exhibits a structure with rich highorder

INTERNET traffic often exhibits a structure with rich highorder IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 21, NO. 11, NOVEMBER 21 1573 Filter Design and Analysis in Frequency Domain for Server Scheduling and Optimization Cheng-hong Xu, Senior Member,

More information

IP Traffic Prediction and Equivalent Bandwidth for DAMA TDMA Protocols

IP Traffic Prediction and Equivalent Bandwidth for DAMA TDMA Protocols IP Traffic and Equivalent Bandwidth for DAMA TDMA Protocols J. Aracil, D. Morato, E. Magaña, M. Izal Universidad Pública de Navarra, 316 Pamplona, SPAIN email:javier.aracil@unavarra.es Abstract The use

More information

ENERGY is a critical resource of wireless devices powered

ENERGY is a critical resource of wireless devices powered IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 8, AUGUST 2010 1187 Maximizing Rewards in Wireless Networks with Energy and Timing Constraints for Periodic Data Streams Jiayu Gong, Student Member, IEEE,

More information

IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 8, NO. 2, APRIL Segment-Based Streaming Media Proxy: Modeling and Optimization

IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 8, NO. 2, APRIL Segment-Based Streaming Media Proxy: Modeling and Optimization IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 8, NO. 2, APRIL 2006 243 Segment-Based Streaming Media Proxy: Modeling Optimization Songqing Chen, Member, IEEE, Bo Shen, Senior Member, IEEE, Susie Wee, Xiaodong

More information

IN the past decade, we have seen an increasing demand for

IN the past decade, we have seen an increasing demand for 964 IEEE TRANSACTIONS ON COMPUTERS, VOL. 54, NO. 8, AUGUST 2005 Robust Processing Rate Allocation for Proportional Slowdown Differentiation on Internet Servers Jianbin Wei, Student Member, IEEE, Xiaobo

More information

Analytic Performance Models for Bounded Queueing Systems

Analytic Performance Models for Bounded Queueing Systems Analytic Performance Models for Bounded Queueing Systems Praveen Krishnamurthy Roger D. Chamberlain Praveen Krishnamurthy and Roger D. Chamberlain, Analytic Performance Models for Bounded Queueing Systems,

More information

Efficient Power Management in Wireless Communication

Efficient Power Management in Wireless Communication Efficient Power Management in Wireless Communication R.Saranya 1, Mrs.J.Meena 2 M.E student,, Department of ECE, P.S.R.College of Engineering, sivakasi, Tamilnadu, India 1 Assistant professor, Department

More information

Effective Capacity: A Wireless Link Model for Support of Quality of Service

Effective Capacity: A Wireless Link Model for Support of Quality of Service 630 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 2, NO. 4, JULY 2003 Effective Capacity: A Wireless Link Model for Support of Quality of Service Dapeng Wu, Student Member, IEEE, and Rohit Negi, Member,

More information

A Path Decomposition Approach for Computing Blocking Probabilities in Wavelength-Routing Networks

A Path Decomposition Approach for Computing Blocking Probabilities in Wavelength-Routing Networks IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 8, NO. 6, DECEMBER 2000 747 A Path Decomposition Approach for Computing Blocking Probabilities in Wavelength-Routing Networks Yuhong Zhu, George N. Rouskas, Member,

More information

A DYNAMIC RESOURCE ALLOCATION STRATEGY FOR SATELLITE COMMUNICATIONS. Eytan Modiano MIT LIDS Cambridge, MA

A DYNAMIC RESOURCE ALLOCATION STRATEGY FOR SATELLITE COMMUNICATIONS. Eytan Modiano MIT LIDS Cambridge, MA A DYNAMIC RESOURCE ALLOCATION STRATEGY FOR SATELLITE COMMUNICATIONS Aradhana Narula-Tam MIT Lincoln Laboratory Lexington, MA Thomas Macdonald MIT Lincoln Laboratory Lexington, MA Eytan Modiano MIT LIDS

More information

Statistical Timing Analysis Using Bounds and Selective Enumeration

Statistical Timing Analysis Using Bounds and Selective Enumeration IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS, VOL. 22, NO. 9, SEPTEMBER 2003 1243 Statistical Timing Analysis Using Bounds and Selective Enumeration Aseem Agarwal, Student

More information

Introduction: Two motivating examples for the analytical approach

Introduction: Two motivating examples for the analytical approach Introduction: Two motivating examples for the analytical approach Hongwei Zhang http://www.cs.wayne.edu/~hzhang Acknowledgement: this lecture is partially based on the slides of Dr. D. Manjunath Outline

More information

LOW-DENSITY PARITY-CHECK (LDPC) codes [1] can

LOW-DENSITY PARITY-CHECK (LDPC) codes [1] can 208 IEEE TRANSACTIONS ON MAGNETICS, VOL 42, NO 2, FEBRUARY 2006 Structured LDPC Codes for High-Density Recording: Large Girth and Low Error Floor J Lu and J M F Moura Department of Electrical and Computer

More information

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 53, NO. 10, OCTOBER

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 53, NO. 10, OCTOBER IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 53, NO. 10, OCTOBER 2007 3413 Relay Networks With Delays Abbas El Gamal, Fellow, IEEE, Navid Hassanpour, and James Mammen, Student Member, IEEE Abstract The

More information

CHAPTER 3 EFFECTIVE ADMISSION CONTROL MECHANISM IN WIRELESS MESH NETWORKS

CHAPTER 3 EFFECTIVE ADMISSION CONTROL MECHANISM IN WIRELESS MESH NETWORKS 28 CHAPTER 3 EFFECTIVE ADMISSION CONTROL MECHANISM IN WIRELESS MESH NETWORKS Introduction Measurement-based scheme, that constantly monitors the network, will incorporate the current network state in the

More information

THE third-generation (and beyond) wireless communication

THE third-generation (and beyond) wireless communication IEEE TRANSACTIONS ON MOBIE COMPUTING, VO. 3, NO., JANUARY-MARCH 2004 73 QoS-Oriented Packet Scheduling for Wireless Multimedia CDMA Communications Vincent Huang and Weihua Zhuang, Senior Member, IEEE Abstract

More information

3 No-Wait Job Shops with Variable Processing Times

3 No-Wait Job Shops with Variable Processing Times 3 No-Wait Job Shops with Variable Processing Times In this chapter we assume that, on top of the classical no-wait job shop setting, we are given a set of processing times for each operation. We may select

More information

Fig. 1. Superframe structure in IEEE

Fig. 1. Superframe structure in IEEE Analyzing the Performance of GTS Allocation Using Markov Model in IEEE 802.15.4 Alladi Ramesh 1,Dr.P.Sumithabhashini 2 1 Dept.of CSE, PETW, Hyderabad 2 Dept.of ECE, PETW, Hyderabad Abstract-In this paper,

More information

IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 15, NO. 1, FEBRUARY

IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 15, NO. 1, FEBRUARY IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 15, NO. 1, FEBRUARY 2007 187 Modeling Best-Effort and FEC Streaming of Scalable Video in Lossy Network Channels Seong-Ryong Kang, Student Member, IEEE, and Dmitri

More information

Performance Analysis of Cell Switching Management Scheme in Wireless Packet Communications

Performance Analysis of Cell Switching Management Scheme in Wireless Packet Communications Performance Analysis of Cell Switching Management Scheme in Wireless Packet Communications Jongho Bang Sirin Tekinay Nirwan Ansari New Jersey Center for Wireless Telecommunications Department of Electrical

More information

Stochastic Control of Path Optimization for Inter-Switch Handoffs in Wireless ATM Networks

Stochastic Control of Path Optimization for Inter-Switch Handoffs in Wireless ATM Networks 336 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 9, NO. 3, JUNE 2001 Stochastic Control of Path Optimization for Inter-Switch Handoffs in Wireless ATM Networks Vincent W. S. Wong, Member, IEEE, Mark E. Lewis,

More information

An Adaptable Neural-Network Model for Recursive Nonlinear Traffic Prediction and Modeling of MPEG Video Sources

An Adaptable Neural-Network Model for Recursive Nonlinear Traffic Prediction and Modeling of MPEG Video Sources 150 IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL 14, NO 1, JANUARY 2003 An Adaptable Neural-Network Model for Recursive Nonlinear Traffic Prediction and Modeling of MPEG Video Sources Anastasios D Doulamis,

More information

DiffServ Architecture: Impact of scheduling on QoS

DiffServ Architecture: Impact of scheduling on QoS DiffServ Architecture: Impact of scheduling on QoS Abstract: Scheduling is one of the most important components in providing a differentiated service at the routers. Due to the varying traffic characteristics

More information

Comparison of Shaping and Buffering for Video Transmission

Comparison of Shaping and Buffering for Video Transmission Comparison of Shaping and Buffering for Video Transmission György Dán and Viktória Fodor Royal Institute of Technology, Department of Microelectronics and Information Technology P.O.Box Electrum 229, SE-16440

More information

36 IEEE TRANSACTIONS ON BROADCASTING, VOL. 54, NO. 1, MARCH 2008

36 IEEE TRANSACTIONS ON BROADCASTING, VOL. 54, NO. 1, MARCH 2008 36 IEEE TRANSACTIONS ON BROADCASTING, VOL. 54, NO. 1, MARCH 2008 Continuous-Time Collaborative Prefetching of Continuous Media Soohyun Oh, Beshan Kulapala, Andréa W. Richa, and Martin Reisslein Abstract

More information

ARELAY network consists of a pair of source and destination

ARELAY network consists of a pair of source and destination 158 IEEE TRANSACTIONS ON INFORMATION THEORY, VOL 55, NO 1, JANUARY 2009 Parity Forwarding for Multiple-Relay Networks Peyman Razaghi, Student Member, IEEE, Wei Yu, Senior Member, IEEE Abstract This paper

More information

654 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 2, MARCH 2006

654 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 2, MARCH 2006 654 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 2, MARCH 2006 Optimal Resource Allocation and Adaptive Call Admission Control for Voice/Data Integrated Cellular Networks Chi Wa Leong, Weihua

More information

Lecture notes on the simplex method September We will present an algorithm to solve linear programs of the form. maximize.

Lecture notes on the simplex method September We will present an algorithm to solve linear programs of the form. maximize. Cornell University, Fall 2017 CS 6820: Algorithms Lecture notes on the simplex method September 2017 1 The Simplex Method We will present an algorithm to solve linear programs of the form maximize subject

More information

Appendix B. Standards-Track TCP Evaluation

Appendix B. Standards-Track TCP Evaluation 215 Appendix B Standards-Track TCP Evaluation In this appendix, I present the results of a study of standards-track TCP error recovery and queue management mechanisms. I consider standards-track TCP error

More information

Week 7: Traffic Models and QoS

Week 7: Traffic Models and QoS Week 7: Traffic Models and QoS Acknowledgement: Some slides are adapted from Computer Networking: A Top Down Approach Featuring the Internet, 2 nd edition, J.F Kurose and K.W. Ross All Rights Reserved,

More information

Assignment 7: TCP and Congestion Control Due the week of October 29/30, 2015

Assignment 7: TCP and Congestion Control Due the week of October 29/30, 2015 Assignment 7: TCP and Congestion Control Due the week of October 29/30, 2015 I d like to complete our exploration of TCP by taking a close look at the topic of congestion control in TCP. To prepare for

More information

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 6, JUNE Jie Luo, Member, IEEE, and Anthony Ephremides, Fellow, IEEE

IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 6, JUNE Jie Luo, Member, IEEE, and Anthony Ephremides, Fellow, IEEE IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 52, NO. 6, JUNE 2006 2593 On the Throughput, Capacity, and Stability Regions of Random Multiple Access Jie Luo, Member, IEEE, and Anthony Ephremides, Fellow,

More information

Scheduling Algorithms to Minimize Session Delays

Scheduling Algorithms to Minimize Session Delays Scheduling Algorithms to Minimize Session Delays Nandita Dukkipati and David Gutierrez A Motivation I INTRODUCTION TCP flows constitute the majority of the traffic volume in the Internet today Most of

More information

End-to-end bandwidth guarantees through fair local spectrum share in wireless ad-hoc networks

End-to-end bandwidth guarantees through fair local spectrum share in wireless ad-hoc networks End-to-end bandwidth guarantees through fair local spectrum share in wireless ad-hoc networks Saswati Sarkar and Leandros Tassiulas 1 Abstract Sharing the locally common spectrum among the links of the

More information

Worst-case Ethernet Network Latency for Shaped Sources

Worst-case Ethernet Network Latency for Shaped Sources Worst-case Ethernet Network Latency for Shaped Sources Max Azarov, SMSC 7th October 2005 Contents For 802.3 ResE study group 1 Worst-case latency theorem 1 1.1 Assumptions.............................

More information

McGill University - Faculty of Engineering Department of Electrical and Computer Engineering

McGill University - Faculty of Engineering Department of Electrical and Computer Engineering McGill University - Faculty of Engineering Department of Electrical and Computer Engineering ECSE 494 Telecommunication Networks Lab Prof. M. Coates Winter 2003 Experiment 5: LAN Operation, Multiple Access

More information

Characterizing Internet Load as a Non-regular Multiplex of TCP Streams

Characterizing Internet Load as a Non-regular Multiplex of TCP Streams Characterizing Internet Load as a Non-regular Multiplex of TCP Streams J. Aracil, D. Morató Dpto. Automática y Computación Universidad Pública de Navarra {javier.aracil,daniel.morato}@unavarra.es http://www.tlm.unavarra.es

More information

On the Performance of Greedy Algorithms in Packet Buffering

On the Performance of Greedy Algorithms in Packet Buffering On the Performance of Greedy Algorithms in Packet Buffering Susanne Albers Ý Markus Schmidt Þ Abstract We study a basic buffer management problem that arises in network switches. Consider input ports,

More information

Optimal Routing and Scheduling in Multihop Wireless Renewable Energy Networks

Optimal Routing and Scheduling in Multihop Wireless Renewable Energy Networks Optimal Routing and Scheduling in Multihop Wireless Renewable Energy Networks ITA 11, San Diego CA, February 2011 MHR. Khouzani, Saswati Sarkar, Koushik Kar UPenn, UPenn, RPI March 23, 2011 Khouzani, Sarkar,

More information

Chapter 1. Introduction

Chapter 1. Introduction Chapter 1 Introduction A Monte Carlo method is a compuational method that uses random numbers to compute (estimate) some quantity of interest. Very often the quantity we want to compute is the mean of

More information

136 Proceedings of the rd International Teletraffic Congress (ITC 2011)

136 Proceedings of the rd International Teletraffic Congress (ITC 2011) Gaussian Approximation of CDN Call Level Traffic Andrzej Bak and Piotr Gajowniczek Institute of Telecommunications Warsaw University of Technology Nowowiejska 5/9, -665 Warsaw, Poland Email: bak@tele.pw.edu.pl

More information

Neuro-fuzzy admission control in mobile communications systems

Neuro-fuzzy admission control in mobile communications systems University of Wollongong Thesis Collections University of Wollongong Thesis Collection University of Wollongong Year 2005 Neuro-fuzzy admission control in mobile communications systems Raad Raad University

More information

Optimizing Cloud Resources for Delivering IPTV. Services through Virtualization

Optimizing Cloud Resources for Delivering IPTV. Services through Virtualization Optimizing Cloud Resources for Delivering IPTV 1 Services through Virtualization Vaneet Aggarwal, Vijay Gopalakrishnan, Rittwik Jana, K. K. Ramakrishnan, Vinay A. Vaishampayan Abstract Virtualized cloud-based

More information

THE capabilities of mobile cellular communications may

THE capabilities of mobile cellular communications may 680 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL. 54, NO. 4, APRIL 2006 Analysis and Modeling of Upstream Throughput in Multihop Packet CDMA Cellular Networks Ali Nabi Zadeh, Member, IEEE, and Bijan Jabbari,

More information

FUTURE communication networks are expected to support

FUTURE communication networks are expected to support 1146 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL 13, NO 5, OCTOBER 2005 A Scalable Approach to the Partition of QoS Requirements in Unicast and Multicast Ariel Orda, Senior Member, IEEE, and Alexander Sprintson,

More information

Wireless Multicast: Theory and Approaches

Wireless Multicast: Theory and Approaches University of Pennsylvania ScholarlyCommons Departmental Papers (ESE) Department of Electrical & Systems Engineering June 2005 Wireless Multicast: Theory Approaches Prasanna Chaporkar University of Pennsylvania

More information

Overview Computer Networking What is QoS? Queuing discipline and scheduling. Traffic Enforcement. Integrated services

Overview Computer Networking What is QoS? Queuing discipline and scheduling. Traffic Enforcement. Integrated services Overview 15-441 15-441 Computer Networking 15-641 Lecture 19 Queue Management and Quality of Service Peter Steenkiste Fall 2016 www.cs.cmu.edu/~prs/15-441-f16 What is QoS? Queuing discipline and scheduling

More information

An Approach to Connection Admission Control in Single-Hop Multiservice Wireless Networks With QoS Requirements

An Approach to Connection Admission Control in Single-Hop Multiservice Wireless Networks With QoS Requirements 1110 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 52, NO. 4, JULY 2003 An Approach to Connection Admission Control in Single-Hop Multiservice Wireless Networks With QoS Requirements Tara Javidi, Member,

More information

Advanced Internet Technologies

Advanced Internet Technologies Advanced Internet Technologies Chapter 3 Performance Modeling Dr.-Ing. Falko Dressler Chair for Computer Networks & Internet Wilhelm-Schickard-Institute for Computer Science University of Tübingen http://net.informatik.uni-tuebingen.de/

More information

A Real-Time Network Simulation Application for Multimedia over IP

A Real-Time Network Simulation Application for Multimedia over IP A Real-Time Simulation Application for Multimedia over IP ABSTRACT This paper details a Secure Voice over IP (SVoIP) development tool, the Simulation Application (Netsim), which provides real-time network

More information

IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 10, NO. 4, AUGUST Energy-Efficient Packet Transmission Over a Wireless Link

IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 10, NO. 4, AUGUST Energy-Efficient Packet Transmission Over a Wireless Link IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 10, NO. 4, AUGUST 2002 487 Energy-Efficient Packet Transmission Over a Wireless Link Elif Uysal-Biyikoglu, Student Member, IEEE, Balaji Prabhakar, Abbas El Gamal,

More information

IN a mobile ad hoc network, nodes move arbitrarily.

IN a mobile ad hoc network, nodes move arbitrarily. IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 5, NO. 6, JUNE 2006 609 Distributed Cache Updating for the Dynamic Source Routing Protocol Xin Yu Abstract On-demand routing protocols use route caches to make

More information

WIRELESS/MOBILE networking is one of the strongest

WIRELESS/MOBILE networking is one of the strongest IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 53, NO. 3, MAY 2004 835 An Integrated Adaptive Bandwidth-Management Framework for QoS-Sensitive Multimedia Cellular Networks Sungwook Kim and Pramod K. Varshney,

More information

THE expanded availability of small wireless computers

THE expanded availability of small wireless computers IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 4, NO. 4, JULY/AUGUST 2005 363 Contention-Aware Admission Control for Ad Hoc Networks Yaling Yang, Student Member, IEEE, and Robin Kravets, Member, IEEE Abstract

More information

Performance Evaluation of Scheduling Mechanisms for Broadband Networks

Performance Evaluation of Scheduling Mechanisms for Broadband Networks Performance Evaluation of Scheduling Mechanisms for Broadband Networks Gayathri Chandrasekaran Master s Thesis Defense The University of Kansas 07.31.2003 Committee: Dr. David W. Petr (Chair) Dr. Joseph

More information

1188 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 13, NO. 5, OCTOBER Wei Sun, Student Member, IEEE, and Kang G. Shin, Fellow, IEEE

1188 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 13, NO. 5, OCTOBER Wei Sun, Student Member, IEEE, and Kang G. Shin, Fellow, IEEE 1188 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 13, NO. 5, OCTOBER 2005 End-to-End Delay Bounds for Traffic Aggregates Under Guaranteed-Rate Scheduling Algorithms Wei Sun, Student Member, IEEE, and Kang

More information

Active Queue Management for Self-Similar Network Traffic

Active Queue Management for Self-Similar Network Traffic Active Queue Management for Self-Similar Network Traffic Farnaz Amin*, Kiarash Mizanain**, and Ghasem Mirjalily*** * Electrical Engineering and computer science Department, Yazd University, farnaz.amin@stu.yazduni.ac.ir

More information

Sustainable Computing: Informatics and Systems 00 (2014) Huangxin Wang

Sustainable Computing: Informatics and Systems 00 (2014) Huangxin Wang Sustainable Computing: Informatics and Systems 00 (2014) 1 17 Sustainable Computing Worst-Case Performance Guarantees of Scheduling Algorithms Maximizing Weighted Throughput in Energy-Harvesting Networks

More information

IN mobile wireless networks, communications typically take

IN mobile wireless networks, communications typically take IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING, VOL 2, NO 2, APRIL 2008 243 Optimal Dynamic Resource Allocation for Multi-Antenna Broadcasting With Heterogeneous Delay-Constrained Traffic Rui Zhang,

More information

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY

IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL. 55, NO. 5, MAY IEEE TRANSACTIONS ON SIGNAL PROCESSING, VOL 55, NO 5, MAY 2007 1911 Game Theoretic Cross-Layer Transmission Policies in Multipacket Reception Wireless Networks Minh Hanh Ngo, Student Member, IEEE, and

More information

FUTURE wireless networks are expected to support multimedia

FUTURE wireless networks are expected to support multimedia 60 IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, VOL. 3, NO. 1, JANUARY 2004 Dynamic Fair Scheduling With QoS Constraints in Multimedia Wideband CDMA Cellular Networks Liang Xu, Member, IEEE, Xuemin (Sherman)

More information

CHAPTER 6 STATISTICAL MODELING OF REAL WORLD CLOUD ENVIRONMENT FOR RELIABILITY AND ITS EFFECT ON ENERGY AND PERFORMANCE

CHAPTER 6 STATISTICAL MODELING OF REAL WORLD CLOUD ENVIRONMENT FOR RELIABILITY AND ITS EFFECT ON ENERGY AND PERFORMANCE 143 CHAPTER 6 STATISTICAL MODELING OF REAL WORLD CLOUD ENVIRONMENT FOR RELIABILITY AND ITS EFFECT ON ENERGY AND PERFORMANCE 6.1 INTRODUCTION This chapter mainly focuses on how to handle the inherent unreliability

More information

SENDING packet-pairs to measure network path characteristics

SENDING packet-pairs to measure network path characteristics IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 24, NO. 12, DECEMBER 2006 2249 Towards a Generalized Stochastic Model of End-to-End Packet-Pair Sampling Xiliang Liu, Member, IEEE, Kaliappa Ravindran,

More information

GreenBag: Energy-efficient Bandwidth Aggregation For Real-time Streaming in Heterogeneous Mobile Wireless Networks

GreenBag: Energy-efficient Bandwidth Aggregation For Real-time Streaming in Heterogeneous Mobile Wireless Networks GreenBag: Energy-efficient Bandwidth Aggregation For Real-time Streaming in Heterogeneous Mobile Wireless Networks Duc Hoang Bui, Kilho Lee, Sangeun Oh, Insik Shin Dept. of Computer Science KAIST, South

More information

CS261: A Second Course in Algorithms Lecture #16: The Traveling Salesman Problem

CS261: A Second Course in Algorithms Lecture #16: The Traveling Salesman Problem CS61: A Second Course in Algorithms Lecture #16: The Traveling Salesman Problem Tim Roughgarden February 5, 016 1 The Traveling Salesman Problem (TSP) In this lecture we study a famous computational problem,

More information

Optimum Scheduling and Memory Management in Input Queued Switches with Finite Buffer Space

Optimum Scheduling and Memory Management in Input Queued Switches with Finite Buffer Space University of Pennsylvania ScholarlyCommons Departmental Papers (ESE) Department of Electrical & Systems Engineering March 23 Optimum Scheduling and Memory Management in Input Queued Switches with Finite

More information

The Computer Journal Advance Access published October 17, 2007

The Computer Journal Advance Access published October 17, 2007 The Computer Journal Advance Access published October 17, 2007 # The Author 2007. Published by Oxford University Press on behalf of The British Computer Society. All rights reserved. For Permissions, please

More information

Optimizing Cloud Resources for Delivering IPTV Services through Virtualization

Optimizing Cloud Resources for Delivering IPTV Services through Virtualization Optimizing Cloud Resources for Delivering IPTV Services through Virtualization Vaneet Aggarwal, Vijay Gopalakrishnan, Rittwik Jana, K. K. Ramakrishnan, Vinay A. Vaishampayan Abstract Virtualized cloud-based

More information

Lecture 2.2 Cubic Splines

Lecture 2.2 Cubic Splines Lecture. Cubic Splines Cubic Spline The equation for a single parametric cubic spline segment is given by 4 i t Bit t t t i (..) where t and t are the parameter values at the beginning and end of the segment.

More information

Quality of Service (QoS)

Quality of Service (QoS) Quality of Service (QoS) The Internet was originally designed for best-effort service without guarantee of predictable performance. Best-effort service is often sufficient for a traffic that is not sensitive

More information

different problems from other networks ITU-T specified restricted initial set Limited number of overhead bits ATM forum Traffic Management

different problems from other networks ITU-T specified restricted initial set Limited number of overhead bits ATM forum Traffic Management Traffic and Congestion Management in ATM 3BA33 David Lewis 3BA33 D.Lewis 2007 1 Traffic Control Objectives Optimise usage of network resources Network is a shared resource Over-utilisation -> congestion

More information

INTEGRATION of data communications services into wireless

INTEGRATION of data communications services into wireless 208 IEEE TRANSACTIONS ON COMMUNICATIONS, VOL 54, NO 2, FEBRUARY 2006 Service Differentiation in Multirate Wireless Networks With Weighted Round-Robin Scheduling and ARQ-Based Error Control Long B Le, Student

More information

Adaptive Linear Prediction of Queues for Reduced Rate Scheduling in Optical Routers

Adaptive Linear Prediction of Queues for Reduced Rate Scheduling in Optical Routers Adaptive Linear Prediction of Queues for Reduced Rate Scheduling in Optical Routers Yang Jiao and Ritesh Madan EE 384Y Final Project Stanford University Abstract This paper describes a switching scheme

More information

Energy-Aware Routing in Wireless Ad-hoc Networks

Energy-Aware Routing in Wireless Ad-hoc Networks Energy-Aware Routing in Wireless Ad-hoc Networks Panagiotis C. Kokkinos Christos A. Papageorgiou Emmanouel A. Varvarigos Abstract In this work we study energy efficient routing strategies for wireless

More information

WIRELESS local area network (WLAN)-based systems are

WIRELESS local area network (WLAN)-based systems are IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 6, NO. 1, JANUARY 2007 1 Optimal Joint Session Admission Control in Integrated WLAN and CDMA Cellular Networks with Vertical Handoff Fei Yu, Member, IEEE, and

More information

Queuing Delay and Achievable Throughput in Random Access Wireless Ad Hoc Networks

Queuing Delay and Achievable Throughput in Random Access Wireless Ad Hoc Networks Queuing Delay and Achievable Throughput in Random Access Wireless Ad Hoc Networks Nabhendra Bisnik and Alhussein Abouzeid Rensselaer Polytechnic Institute Troy, NY bisnin@rpi.edu, abouzeid@ecse.rpi.edu

More information

MPEG VIDEO TRAFFIC MODELS: SEQUENTIALLY MODULATED SELF-SIMILAR PROCESSES

MPEG VIDEO TRAFFIC MODELS: SEQUENTIALLY MODULATED SELF-SIMILAR PROCESSES MPEG VIDEO TRAFFIC MODELS: SEQUENTIALLY MODULATED SELF-SIMILAR PROCESSES Hai Liu, Nirwan Ansari, and Yun Q. Shi New Jersey Center for Wireless Telecommunications Department of Electrical and Computer Engineering

More information

Lecture 5: Performance Analysis I

Lecture 5: Performance Analysis I CS 6323 : Modeling and Inference Lecture 5: Performance Analysis I Prof. Gregory Provan Department of Computer Science University College Cork Slides: Based on M. Yin (Performability Analysis) Overview

More information

Practical Lazy Scheduling in Wireless Sensor Networks. Ramana Rao Kompella and Alex C. Snoeren

Practical Lazy Scheduling in Wireless Sensor Networks. Ramana Rao Kompella and Alex C. Snoeren Practical Lazy Scheduling in Wireless Sensor Networks Ramana Rao Kompella and Alex C. Snoeren Distributed Rate Adaptation Problem: In wireless networks (e.g., sensor nets, 802.11) radios consume significant

More information

On the Max Coloring Problem

On the Max Coloring Problem On the Max Coloring Problem Leah Epstein Asaf Levin May 22, 2010 Abstract We consider max coloring on hereditary graph classes. The problem is defined as follows. Given a graph G = (V, E) and positive

More information

CHAPTER 5. QoS RPOVISIONING THROUGH EFFECTIVE RESOURCE ALLOCATION

CHAPTER 5. QoS RPOVISIONING THROUGH EFFECTIVE RESOURCE ALLOCATION CHAPTER 5 QoS RPOVISIONING THROUGH EFFECTIVE RESOURCE ALLOCATION 5.1 PRINCIPLE OF RRM The success of mobile communication systems and the need for better QoS, has led to the development of 3G mobile systems

More information

Chapter -5 QUALITY OF SERVICE (QOS) PLATFORM DESIGN FOR REAL TIME MULTIMEDIA APPLICATIONS

Chapter -5 QUALITY OF SERVICE (QOS) PLATFORM DESIGN FOR REAL TIME MULTIMEDIA APPLICATIONS Chapter -5 QUALITY OF SERVICE (QOS) PLATFORM DESIGN FOR REAL TIME MULTIMEDIA APPLICATIONS Chapter 5 QUALITY OF SERVICE (QOS) PLATFORM DESIGN FOR REAL TIME MULTIMEDIA APPLICATIONS 5.1 Introduction For successful

More information

206 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 16, NO. 1, FEBRUARY The RGA arbitration can also start from the output side like in DRR [13] and

206 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 16, NO. 1, FEBRUARY The RGA arbitration can also start from the output side like in DRR [13] and 206 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 16, NO. 1, FEBRUARY 2008 Matching From the First Iteration: An Iterative Switching Algorithm for an Input Queued Switch Saad Mneimneh Abstract An iterative

More information

Network Topology Control and Routing under Interface Constraints by Link Evaluation

Network Topology Control and Routing under Interface Constraints by Link Evaluation Network Topology Control and Routing under Interface Constraints by Link Evaluation Mehdi Kalantari Phone: 301 405 8841, Email: mehkalan@eng.umd.edu Abhishek Kashyap Phone: 301 405 8843, Email: kashyap@eng.umd.edu

More information

Performance Study of Routing Algorithms for LEO Satellite Constellations

Performance Study of Routing Algorithms for LEO Satellite Constellations Performance Study of Routing Algorithms for LEO Satellite Constellations Ioannis Gragopoulos, Evangelos Papapetrou, Fotini-Niovi Pavlidou Aristotle University of Thessaloniki, School of Engineering Dept.

More information

Markov Chains and Multiaccess Protocols: An. Introduction

Markov Chains and Multiaccess Protocols: An. Introduction Markov Chains and Multiaccess Protocols: An Introduction Laila Daniel and Krishnan Narayanan April 8, 2012 Outline of the talk Introduction to Markov Chain applications in Communication and Computer Science

More information