Idle-Slot Recycling in a Collision-Free Real-Time MAC Protocol

Transcription

1 Idle-Slot ecycling in a Collision-Free eal-time MAC Protocol Ming Zhang Ying Jian Liang Zhang Shigang Chen Department of Computer & Information Science & Engineering University of Florida {mzhang, yjian, lzhang, sgchen}@cise.ufl.edu Abstract In wireless sensor networs (WSNs), timeliness is one of the most challenging problems for critical applications. Caccamo et al. designed a collision-free real-time MAC protocol for sensor networs by adopting the cellular structure in traditional telecommunication networs. They assumed that sensors are organized into many rigid hexagon cells and a router node is at the center of each cell to transmit inter-cell pacets. By exploiting FDMA and TDMA, transmission collisions are avoided and real-time guarantees are provided. However, we observe that for inter-cell communication, idles caused by the TDMA-based scheduling will degrade the throughput and delay performance of the entire networ, especially for some real applications in which data flows have traffic- partiality characteristics. In the worst case, five sixths of inter-cell bandwidth will be wasted. In this paper, we propose four idle- recycling algorithms to improve channel utilization. Simulation results show that our proposed algorithms can greatly increase networ throughput and decrease pacet transmission delay, while the collision-free and real-time qualities of Caccamo s protocol are still retained. I. INTODUCTION In recent years, wireless sensor networs (WSNs) have been identified as a promising technology for a wide variety of applications [1], such as environment monitoring and military surveillance, etc. Due to the tight interaction with physical environment, these applications are expected to have implicit/explicit real-time requirements [2]. One promising collision-free real-time MAC protocol is proposed by Caccamo et al. in [3]. The protocol adopts the cellular structure in traditional telecommunication networs [4], and it assumes a rigid deployment structure with a node (called router) at the center of each hexagon cell. Using FDMA (Frequency Division Multiple Access), different communication channels are assigned to neighboring cells; seven channels are needed for the whole networ. Since each cell is assigned a channel that is different from its six adjacent cells, all cells can transmit simultaneously without conflicts. Time is also divided into periodical intra-cell s and intercell s. During an intra-cell, sensors inside each cell use the same local channel to transmit pacets. Traffic across Acnowledgement: This wor was supported in part by the US National Science Foundation under grant CNS and by the University of Florida through the esearch Opportunity Fund Award under project no F E A A D B C B C Fig. 1. CH.3 CH.7 CH.6 CH.1 CH.5 CH.3 CH.4 Inter-cell communication In A at time 0 Collision-Free eal-time MAC Protocol[3] D cells is relayed by the routers at the center of each cell. A router is a more capable node equipped with both a transmitter and a receiver, such that it can send a pacet (in the local channel) and receive a pacet (in a neighbor s channel) simultaneously. The inter-cell communication of the whole networ is synchronized by a TDMA (Time Division Multiple Access) scheduling policy, in which all routers simultaneously transmit in the same for each inter-cell time. There are six s (A-F), and they are chosen one by one in consecutive inter-cell time s. For example, as illustrated in Fig. 1, at time 0, all routers transmit in A. Under this protocol, it is guaranteed that no collision will occur in intra-cell and inter-cell communication, and a hard bound for pacet forwarding time can be given. However, we may observe that the inter-cell s are equally and periodically assigned to the six s, while for many applications, it is very liely that during a certain period of time, the local traffic only goes along a few s. We call this characteristic as traffic- partiality. It is easy to see that for data flows with traffic- partiality, a lot of inter-cell s will be idle. In the worst case, five sixths of inter-cell bandwidth will be wasted. Previously, there are two types of TDMA idle s reuse. The first type is on spacial time reuse, which tries to assign the same time to maximal number of geographically separated nodes so that interference is minimized and system /08/$ IEEE. 1 This full text paper was peer reviewed at the of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

2 throughput is improved [5], [6], [7]. The other type, which is similar to the problem that we are solving here, is focused on reusing idle time s that are caused by unbaclogged flows [8], [9], [10]. However the existing solutions are designed for some specific TDMA systems and cannot be directly applied to the cellular structured networs studied in this paper. In this paper, we propose four algorithms (-IS, MU- IS, I-IS, ID-IS) to reuse the inter-cell idle-s. When a time becomes idle, the four algorithms employ different strategies to select a to send (receive) a pacet, by which the idle- is expected to be reused. In -IS, as the simplest way, the is randomly selected. In MU- IS, historical transmission information is exploited, and the of the most recent successful transmission is piced. By piggybacing buffer status in pacets, I-IS and ID-IS mae use of future traffic information of neighboring routers to better choose the recycling. Our proposed algorithms are able to greatly improve networ throughput and reduce pacet transmission delay. ID-IS can also allocate the extra bandwidth (gained by idle- recycling) among flows fairly. For all the algorithms, the collision-free and real-time qualities of the original protocol are still retained. We evaluate the proposed algorithms through simulations. The rest of the paper is organized as follows: The problem is defined in Section II. In Section III, four opportunistic idle- recycling algorithms are presented and compared in details. In Section IV, the effectiveness of the proposed algorithms is evaluated through simulations. Finally, we conclude the paper in Section V. II. POBLEM DEFINITION As we have mentioned, in a cellular structured sensor networ with the original protocol given in [3], a considerable amount of inter-cell time s can be idle and a large portion of bandwidth is wasted. We try to solve this problem by designing algorithms to properly reuse the idle-s during inter-cell communication. In this section we tae a closer observation on the occurrences of idle-s and the feasibility of idle- reuse. We also introduce the major concerns that must be addressed when designing idle- reuse algorithms. To simplify description, in the following sections of this paper, if not explicitly specified, the terms time and idle- always denote inter-cell time and inter-cell idle-. The traffic- partiality characteristic is very common for sensor networs, where most data pacets are forwarded from data sources to one or several BSs (base stations). The occurrence of idle-s is a direct result of this traffic- partiality characteristic. For example, suppose that a BS is located to the A of a router, most traffic relayed by would be heading to A. Thus during the time s that are assigned to the other five s, will not receive nor send any pacets. This is obviously a waste of channel capacity, as during these idles and its neighboring routers indeed can eep sending Fig. 2. Fairness in Idle-Slot ecycling and receiving pacets along A. There are two types of idle-s for a router. If during a time a router does not have any pacets to send to the pre-assigned x, wecallitasasending idle of. Similarly, if during a time there is no any pacets from pre-assigned x to, we call it as a receiving idle- of. For two neighboring routers to reuse an idle-, the sending idle- of the sender and the receiving idle- of the receiver must be matched together, by which we mean that the sender must choose to use the idle- to transmit pacets to the receiver, and the receiver must also choose to use the idle- to receive pacets from the sender. It is worth noting that due to the cellular structure and the seven distinct channels, idle- reuse will not cause transmission collisions. For example, as displayed in Fig. 2, even mistaenly reuses its sending idle- to send a pacet to in A, the transmission between and will not be interrupted because and use the same channel while uses a different one. To effectively and decently reuse idle-s, we need to answer the following three questions: (1) Do routers need to exchange idle- information? (2) How do routers maximize the probability of successful idle- reuse? (3) How do routers reuse these idle-s fairly? For question (1), it depends on how effectively we want to reuse the idle-s. By exchanging idle- information among neighboring routers, it is expected that better performance can be achieved. We can piggybac idle- information in data and ACK pacets, and as the attached information can be as short as several bits, the introduced overhead is acceptable. For question (2), the most challenging problem for idle- reuse is that, when two routers have sending and receiving idle- respectively, how they can have a high probability to pic each other to reuse this idle-. Because they both have five other neighbors, the probability of successful reuse can be as low as 4%, if they just flip a coin. For question (3), fairness is another concern, which is important for a networ to provide better service. As shown in Fig. 2, if router and both have many data pacets to send to, should try to fairly allocate its receiving idle-s between them. We will further discuss these questions and give our solutions in the next section /08/$ IEEE. 2 This full text paper was peer reviewed at the of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

3 III. IDLE-SLOT ECYCLING In this Section, four idle- recycling algorithms will be proposed and compared. The basic idea is that, some idle- reuse agreements will be set up between routers to increase the probability of successful idle- reuse. The fairness issue will also be addressed. A. andom Idle-Slot ecycling (-IS) -ISs a very simple way to reuse idle-s: at an arbitrary time, if a router has a sending idle- (receiving idle-), it randomly pics up one among other five s to send (receive) a pacet. The problem of this algorithm is also straightforward: the probability of successful idle- reuse can be very small. Suppose during the same time two neighboring routers and have one sending idle- and one receiving idle respectively, because both and have five other neighbors, the probability for them to pic up each other to reuse their idle-s is as low as 4%. B. Most ecently Used Idle-Slot ecycling (MU-IS) In a typical multi-hop wireless sensor networ, a data source periodically reports collected data to BSs. So once a router node successfully receives a pacet from its neighbor, may anticipate that there will be more following pacets from. Based on this observation, we designed the MU-IS, a quite simple yet efficient idle- recycling algorithm. The basic idea is that once router successfully transmits a pacet to its neighbor,anopportunistic preemptive idle reuse agreement (OP-agreement) is immediately set up between and, and then if has more sending idles, it will try to send pacets to and also will try to receive pacets from at its receiving idle-s. We call this idle- reuse agreement opportunistic and preemptive because these agreements between routers are only based on the current transmission and a guess for future traffic. Besides, these agreements might be set up and preempted frequently and dynamically according to the most recent successful pacet transmissions. For example, as shown in Fig. 3, in A, the router transmits a pacet to and then an OP-agreement is established between them. In the next B, if receives a pacet from its another neighbor, a new OP-agreement is immediately set up between and without regard to the previous OPagreement between and. So in the following C, if has a receiving idle-, it will try to reuse this idle- with instead of. Due to the traffic- partiality nature of data flows, MU-IS wors much better than -IS. However, since OP-agreement is set up opportunistically without accurate information for future traffic, it may not always wor well. As shown in Fig. 3, when releases its OP-agreement with and establishes a new one with, it actually does not now (a) A (b) B (c) C Fig. 3. Opportunistic Preemptive Idle- euse Agreement whether has more pacets for it or not in the near future. If has no more pacets or has no sending idle- in the next C, then the receiving idle- of will be wasted and at the same time the sending idle- of can not be utilized. Another problem is fairness, since the OP-agreement is preemptive, some nodes might have more chances to reuse idle-s than others. For example, as illustrated in Fig. 3, suppose A is followed by B, if both and have many pacets to send to, will always be able to preempt the OP-agreement from, such that may send more pacets than. Therefore, we will propose another two algorithms to further enhance the performance of MU-ISn the following subsections. C. Informed Idle-Slot ecycling (I-IS) The main differences between MU-IS and I-IS are that idle- reuse agreement is set up according to more accurate future traffic information and each agreement will be ept for a certain amount of time and cannot be arbitrarily preempted. I-ISs outlined as follows: (1) Each idle- reuse agreement has a reuse point (P). If P is greater than 0, the agreement is called an informed coordinated idle- reuse agreement (IC-agreement) and it cannot be preempted by others. On the other hand, if P is less than 1, the agreement is called an informed preemptive idle- reuse agreement (IP-agreement), which means new agreement can be set up to replace it. (2) When a router sends a data pacet to,if has more pacets for and does not have any existing ICagreement with others, will attach its buffer status and an idle- reuse request (I-equest) in the data pacet. (3) Once receives an I-equest from,if has receiving idle-s and has no existing IP-agreement with nor IC-agreement with others, sets up an IC-agreement with (P is initialized to a positive number, e.g. 10 in our simulations). The agreement set up confirmation will be attached in the ACK pacet to. (4) Suppose there is an IC-agreement (IP-agreement) between and. When has a sending idle-, will try to send a pacet to. On the other hand, will also try to receive a pacet from at its receiving idle-s. If they succeed to reuse an idle-, the P of the IC-agreement (IPagreement) is deducted by a small number (e.g. 1), otherwise P is deducted by a larger number (e.g. 3). When P of an /08/$ IEEE. 3 This full text paper was peer reviewed at the of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

4 IC-agreement is decreased to less than 1, the IC-agreement is converted to an IP-agreement. Since the future traffic information is exploited, the problem caused by the opportunism in MU-ISs relieved. Moreover, the idle- reuse chances are shared among neighbors more fairly by introducing two types of agreements. For example, as shown in Fig. 3, we pointed out that in MU- IS, will always preempt the OP-agreement from.but in I-IS, the IC-agreement between and cannot be preempted by before it turns to an IP-agreement. D. Informed Directional Idle-Slot ecycling (ID-IS) For the above algorithms, if two routers set up an agreement, it will be applied to all their six s and the al reuse is not taen into account. For example, as shown in Fig. 4, suppose that has pacets for both and, and has an idle- reuse agreement with.in E, can reuse its sending idle- with, while in A, cannot successfully transmit to because has pacets for. As we can see, in fact, in A, should reuse with since has no any pacet to receive. So, if the idle- reuse agreements can be set up by s, we can expect to reuse more idle-s. Thus, based on I-IS, ID-ISs proposed to reuse idles by s. The ID-ISs outlined as follows: (1) As in I-IS, there are two types of idle- reuse agreements, IP-agreement and IC-agreement, and each agreement has a P. For each router, it has six sending s and six receiving s. At the beginning, all these sending and receiving s are mared as unreserved. An agreement can be set up between one sending and one receiving of two routers. (2) When sends a data pacet to, if has more pacets for and has some unreserved sending s, will attach its buffer status, unreserved sending s and an I-equest in the data pacet. (3) Once receives an I-equest from,if has unreserved receiving s, sets up IC-agreements with in matched s (i.e. in these s, and have unreserved receiving and sending s respectively). Then, mars the corresponding receiving s as reserved. But if already has an existing IPagreement with in a certain, just eeps the IP-agreement. The agreement set up confirmation will be sent bac to via ACK. Upon receiving the ACK, sets up agreements and mars its sending s accordingly. (4) Suppose and have an IC-agreement (IPagreement) in A. When has a sending idle- in A, will try to send a pacet to.onthe other hand, when has a receiving idle- in A, will try to receive a pacet from. If they succeed to reuse an idle-, the P of the IC-agreement (IP-agreement) is deduced by a small number, otherwise P is decreased by a larger number. When the P of an IC-agreement is less than (a) E (b) A (c) B Fig. 4. Informed Directional Idle-Slot ecycling 1, the IC-agreement is converted to an IP-agreement and the corresponding is mared as unreserved. IV. SIMULATION We evaluate our proposed idle- recycling algorithms (-IS, MU-IS, I-IS, and ID-IS) through two sets of simulations. We first study their throughput and delay performance, and then we focus on the flow rate fairness they attained. In our simulations, there are 200 router nodes (cells) deployed in a squared area. Each inter-cell and intracell time is 0.01 second. Only inter-cell communications are simulated and during each inter-cell time one pacet can be transmitted. Every pacet has a fixed length of 512 bytes. Simulation results show that all the three algorithms (MU-IS, I-IS, and ID-IS) can greatly improve networ throughput and decrease pacet transmission delay, where ID- IS gives the highest throughput. In addition, ID-IS can achieve better fairness than MU-IS and I-IS. In this section, the original protocol in [3] is used as a baseline for comparison purpose, and it is referred as no recycling. In the first set of simulations, there are two BSs (base stations), one of which is located at the northwest corner of the networ and the other one is located at the northeast corner. 10 out of the 200 nodes are randomly selected as data sources and they periodically send pacets to one of the two BSs. We run the simulation for 200 seconds. Fig. 5 compares the networ throughput achieved by different algorithms when varying the sending rate of each data source. We can see that with the sending rate increasing the networ quicly gets congested if no idle- reuse is exploited. In other words, the networ bandwidth is not well utilized. -IS achieves higher throughput but the improvement is limited. As we expected, the other three algorithms (MU- IS, I-IS, and ID-IS) are able to improve the networ throughput dramatically. Among the three algorithms, ID-IS gets the best performance and I-ISs also better than MU- IS, because informed algorithms can obtain more accurate information about future traffic, by which they can better reuse idle-s accordingly. We study the delay performance of the algorithms when the networ traffic is light, as pacets will be dropped when congestion happens. Fig. 6 shows the simulation results. Because idle-s are well utilized, the pacet transmission delays of MU-IS, I-IS and ID-IS are much smaller than /08/$ IEEE. 4 This full text paper was peer reviewed at the of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.

5 Fig. 5. Fig. 6. Delivered pacets per second Delay in seconds no recycling -IS MU-IS I-IS ID-IS Sending rate (pacets per second) Networ throughput with different source sending rates no recycling -IS MU-IS I-IS ID-IS Sending rate (pacets per second) Transmission delay with different source sending rates those of the other two algorithms. The three curves (MU- IS, I-IS, and ID-IS) stic together because they all wor very well and approach the optimal delay performance. In the next set of simulations, we evaluate the fairness property. We set up two data flows, one of which goes from north to south and the other one goes from west to east, and they cross each other at the center of the networ, where the crossing node becomes the bottlenec. The sending rates of the two flows are set to high enough, such that they are always baclogged at the bottlenec. We observe the effective throughput of the two flows when they are competing with each other. From Table I, no recycling and -IS achieve near perfect fairness because they only allocate the same minimum bandwidth to each of the two flows and leave the rest bandwidth (idle-s) unused, or very limitedly reused. As the other three algorithms (MU-IS, I-IS, and ID-IS) exploit idle-s, the throughput is improved. Especially, ID- TABLE I THOUGHPUT (PACKETS PE SECOND) FAINESS FO TWO COSSING FLOWS flow 1 flow 2 no recycling IS MU-IS I-IS ID-IS IS achieves the best fairness because the idle- recycling in ID-ISs more elegantly coordinated. V. CONCLUSION The MAC protocol proposed in [3] for cellular structured sensor networs gives a promising solution for applications with real-time requirements. By eliminating transmission collisions, it saves energy and can potentially increase networ throughput. However, when traffic- partiality exists, which is very common in sensor networs, a lot of time s become idle and the networ capacity is not fully utilized. In this paper, we proposed four idle- recycling algorithms (-IS, MU-IS, I-IS, ID-IS) that can greatly improve networ throughput and reduce pacet transmission delay. Among the four algorithms, -IS and MU-IS do not need any auxiliary information, and MU-IS already performs quite well for most application scenarios. By maing use of the traffic information of neighboring cells, I-IS and ID-ISmprove the networ performance further. ID- IS achieves the best throughput, while it is able to allocate bandwidth among data flows fairly. Moreover, for all the four algorithms, the collision-free and real-time qualities of the original protocol are retained. The performance of the algorithms is evaluated through simulations. Our future wor will be focused on extending the algorithms to more general TDMA schemes without cellular structures. EFEENCES [1] I. F. Ayildiz, W. Su, Y. Sanarasubramaniam, and E. Cayirci, Wireless sensor networs:a survey, International Journal of Computer and Telecommunications Networing Elsevier, 38(4): , Mar [2] J. A. Stanovic, T. F. Abdelzaher, C. Lu, L. Sha, and J. C. Hou, ealtime communication and coordination in embedded sensor networs, In Proc. of the IEEE, Volume 91, Pages , Jul [3] M. Caccamo, L. Y. Zhang, L. Sha, and G. Buttazzo, An implicit prioritized access protocol for wireless sensor networs, in Proc. of IEEE eal-time Systems Sysmposium 02, Dec [4] D. J. Goodman, Cellular pacet communications, IEEE Transactions on Communications, 38(8), Aug [5] I. Chlamtac and S. Kutten, A spatial reuse tdma/fdma for mobile multihop radio networs, in Proc. IEEE INFOCOM Conf., pp , Mar [6] N. Funabii and Y. Taefuji, A parallel algorithm for broadcast scheduling problems in pacet radio networs, IEEE Trans. Commun., 41(6),pp , Jun [7] I. Cidon and M. Sidi, Distributed assignment algorithms for multihop pacet radio networ, IEEE Trans. Comput., 10(38), pp , [8]. Mnagharam, M. Demirhan,. ajumar, and D. aychaudhuri, Size matters: size-based scheduling for mpeg-4 over wireless channels, SPIE conference on Multimedia Computing and Networing 2004, Jan [9] E. Kwon, D. Hwang, and J. Lim, An idle time reuse scheme for ieee high-rate wireless personal area networs, IEEE Vehicular Technology Conference (VTC)., Vol.2, pp , Sept [10] S. hee, K. Chung, Y. Kim, W. Yoon, and K. Chang, An applicationaware mac scheme for ieee high-rate wpan, IEEE Wireless Communications and Networing Conf., (WCNC), vol.2, pp , Mar /08/$ IEEE. 5 This full text paper was peer reviewed at the of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2008 proceedings.