Modelling OBEX over IrDA Protocol Stack

Transcription

1 Modelling OBEX over IrDA Protocol Stack Pi Huang and A.C. Boucouvalas Multimedia Communications Research Group School of Design, Engineering and Computing Bournemouth University, Fern Barrow, Poole, BH12 5BB, U.K. {phuang, Abstract OBEX, developed by IrDA, is a session layer protocol adopted as the framework for wireless object exchange for wireless transports including indoor Infrared and Bluetooth. In this paper, a mathematical model is developed which leads to derivation of the OBEX throughput over the IrDA protocol stack. The model allows the evaluation of the impact of the IrLAP link layer parameters, OBEX packet size and turnaround time on system throughput for various data rates and in the presence of bit errors. The equations for obtaining the optimum link layer parameters and the OBEX packet size are also derived for maximum OBEX throughput. The results show significant improvement on OBEX performance using optimised parameters. Index Terms OBEX, IrDA, Bluetooth, WPA, optimisation 1. ITRODUCTIO In the field of Wireless Personal Area etwork (WPA) communications, IrDA (Infrared Data Association) and Bluetooth are two of the most widely adopted technologies. A large number of portable devices on the market today have been equipped with IrDA ports and more recently Bluetooth chipsets for their wireless communication needs. These devices range widely from mobile phones and digital cameras to portable computers and printers, [1] and [2]. In recent years, 1

2 there have been many studies on the design and performance of physical and link layers of IrDA and Bluetooth. In contrast there have not been enough studies to examine the important interaction between the lower and higher layers. OBEX is a compact, efficient session layer protocol that enables a wide range of devices to exchange data in a simple and spontaneous manner. OBEX has been defined by members of IrDA for the purpose of interconnecting a wide range of devices that support IrDA protocols [3]. It is not, however, limited in use to only the IrDA environment. OBEX has been adopted by other wireless technology transports, including Bluetooth, as the framework for wireless object exchange. In this paper, we are going to undertake a detailed study of the performance of OBEX protocol operating on top of the IrDA protocol stack. Many IrDA link layer performance improvements and evaluations have been undertaken recently to address different issues including minimum link turnaround time [4] and processing speed [5]. In [6], optimum link parameters used to maximize the link throughput are also presented. All previous evaluations and optimisations focus on the link performance by always assuming an infinite size of upper layer packet ready to transmit. In reality, however upper layers offer only finite size packets to lower layers. OBEX is an example of such a high layer protocol operating similar to a stop and wait protocol [3] and has a finite maximum packet size of 512Kbit delivered to lower layers. It is therefore of interest to examine if the previous optimisation studies which centred on link layer performance only, are also adequate for the upper layers. In [7], a study of OBEX is carried out to search for a suitable OBEX packet size to optimize the system performance. However, no systematic OBEX analysis has yet been developed far. By considering the lower layers and including presence of errors, in this paper, we investigate the OBEX protocol behaviour in detail and derive a comprehensive and more realistic model for OBEX. The paper is organized as follows: first, briefly description of the IrDA protocol stack and OBEX is given. Then, a mathematical model is derived for OBEX which allows derivation of throughput taking into account the lower IrDA protocol stack. Based on this model, the impact of OBEX packet size and IrLAP (IrDA link layer) parameters on the system throughput is studied in the presence of error. In an effort to improve the performance of OBEX when line BER is variable, the OBEX and link layer parameters are optimized which allow maximum throughput for any BER. Analytical results are presented which are also verified by comparing with the results derived from the exact algorithm. Finally we study the effects of maximum OBEX packet size and OBEX turnaround time on throughput for different link data rates. 2

3 2. IRDA PROTOCOL STACK The IrDA protocol stack is a layered set of protocols running on devices for short distance wireless indoor point-to-point infrared communications. The rapid development of IrDA standards has made implementation on many computer platforms possible and more recently IrDA has also become available into many embedded applications. A brief description of the IrDA protocol stack is as follows: A. Physical Layer: The Physical Layer includes the optical transceiver, and deals with shaping and other characteristics of infrared signals including the encoding of data bits. Framing data such as begin and end of frame flags (BOFs and EOFs) and cyclic redundancy checks (CRCs) are also considered to be part of the physical layer although often implemented in software. The maximum data rate defined by IrDA is up to 16 Mbps [8], [9]. B. IrLAP (Link Access Protocol): IrLAP is the IrDA link layer protocol. It is based on High-Level Data Link Control (HDLC) and Synchronous Data Link Control (SDLC) with extensions for some unique characteristics of infrared communications [10]. IrLAP provides simplex reliable data transfer using the following mechanisms: Low-level flow control Error detection Retransmission. By dealing with reliable data transfer at low level, upper layers are free from this concern and can be assured correct delivery. IrLAP transmits data as frames and windows subject to go-back- (GB) error recovery. When the transmitter completes a full window transmission, it sets the Poll (P) bit in the last data frame to signal a request for an acknowledgement from the receiver. Referring to standards [9] and [10], the window size and frame size range from and 512 bit bit respectively. According to [10], the 7 frames IrLAP window transmission model is illustrated in Fig. 1. Fig. 1(a) represents error free data transmission. Fig. 1(b) illustrates the case when there is an error in frame #2 and Fig. 1(c) shows errors in frames #2 and at the last frame of the window, #6 where the P bit was set and therefore lost. 3

4 t I t w P t w (a) t ta tack t s S7F t ack P (b) S2F t w t Fout t s t ack P S0P 2 (c) S2F Fig. 1. (a) Error free transmission of an IrLAP window (b) Retransmission frames due to error frame at I=3 (c) Retransmitted frames and F-timer delay due frame error at I=3 and I=7. C. IrLMP (Link Management Protocol): IrLMP is a required IrDA layer [11], which depends upon the reliable connection and negotiated performance provided by the IrLAP layer. It provides link management multiplexing (LM MUX) which allows multiple IrLMP clients to run over a single IrLAP link, as well as offering a higher-level discovery, address conflict resolution and information access service (IAS). After the connection initial negotiation, IrLMP adds two bytes header to each of the upper layer packet as the link management information. The header provides the Link Service Access Point Selector (LSAP SEL) number of itself and its peer for distinguishing different IrLMP connections between two devices. D. TinyTP (Tiny Transport Protocol): TinyTP is an optional IrDA layer, although it is so important that it should generally be considered a required layer [12]. TinyTP is a flow control mechanism for use with IrLMP. Whilst IrLAP provides a flow-control mechanism between peer IrLAP entities, the introduction of multiplexed channels above IrLAP by IrLMP LM-MUX introduces a problem. Reliance on IrLAP to provide flow-control for a multiplexed channel can result in deadlocks if data flowing in one multiplexed channel is dependent on data flowing in an adjacent multiplexed channel. Conversely, if inbound data on a multiplexed channel cannot be consumed and the underlying IrLAP 4

5 connection cannot be flow-controlled off due to the possibility of deadlock, inbound data (freshly arrived or buffered) must be discarded in the event of buffer exhaustion. To reintroduce reliable data transmission, TinyTP provides two functions on top of IrLMP: Flow control on a per-lmp-connection (per-channel) basis Segmentation and reassembly (SAR). For TinyTP, the entire data packet from upper layers (e.g. OBEX in this case) can be segmented and reassembled in Service Data Units (SDU), and the maximum SDU size is negotiated when the TinyTP/IrLMP connection is first made. The maximum size of the TinyTP segment is limited by the maximum IrLAP frame size, which is determined by IrLAP negotiation. If the communicating peers have large enough buffer size and short propagation delay between them, which is the case for OBEX applications, there is no need for TinyTP to perform flow control. In this paper, we assumed that the IrDA communication peers have sufficient buffer size to accommodate the incoming segments. TinyTP adds one byte of information overhead to each upper layer packet to perform its task. E. OBEX (Object Exchange Protocol): OBEX is a session protocol and can best be described as a binary HTTP protocol, which operates on top of a reliable connection with a simple request and response paradigm. However, OBEX works for many very useful devices that support IrDA or Bluetooth communications but cannot afford the substantial resources required for an HTTP server, and it also targets devices with different usage models that require connection to the Web. OBEX is just like HTTP to serve as a compact final hop to a device. A major use of OBEX is as a Push or Pull application, allowing rapid and ubiquitous communications among portable devices in dynamic environments. For instance, a laptop user pushes a file to another laptop or PDA; a digital camera pushes its pictures into a film development kiosk, or if lost can be queried (pulled) for the electronic business card of its owner. However, OBEX is not limited to quick connect-transfer-disconnect scenarios. It also allows sessions in which transfers take place over a period of time, maintaining the connection even when it is idle. OBEX follows a client/server request-response (stop and wait) paradigm for the conversation format [3]. The terms client and server refer to the originator/receiver of the OBEX connection, not necessarily the one who originated the low level IrLAP connection. Requests are issued by the 5

6 client (the party that initiates the OBEX connection). Once a request is issued, the client waits for a response from the server before issuing another request. The request/response pair is referred to as an operation. PUT and GET are the two types of operations used in OBEX. As the name indicates, the PUT operation sends one object from the client to the server, while the GET operation requests that the server return an object to the client. The maximum and minimum length for both request and response packets are 512Kbit and 2048bit respectively [3]. Fig.2 illustrates OBEX in the process of packetising a large object for transmission when OBEX is in the PUT operation. The initial OBEX request packet (first packet) will typically, although not strictly required to do so, have certain headers. o specific requirements are imposed to the first packet, we assume that first packet includes name, length and body header. The connection-oriented session allows capabilities and information to be exchanged just once at the start of the connection, and allows state information to be kept. The subsequent packets therefore only have to give the overhead information of the packet length field and the body header length field. Object 1 byte 2 bytes Response code Response length OBEX Packets REQ 1 RES 1 REQ 2 RES 2 REQ 3 RES 3 REQ RES 1 byte 2 bytes 1 byte 2 bytes l 1 byte 4 bytes 1 byte 2 bytes n bytes 1 byte 2 bytes 1 byte 2 bytes n bytes OpCode Packet Length ame Header ame Length ame Length header Object Length Body Header Body Length OBEX Turnaround payload OpCode Packet Length Fig. 2. OBEX object packetization Body Header Body Length payload 3. MATHEMATICAL MODEL For the purpose of developing the mathematical model, the following assumptions are made: The OBEX packets are sent in the OBEX PUT operation mode. Only one OBEX application is active in the sender, IrDA link therefore only occupies one IrLMP LSAP assigned for this application. 6

7 Only the connected OBEX packet (not the first packet) is considered. Packet processing time and propagation delay are small enough to be ignored. The buffer of the peer is large enough to accommodate the incoming OBEX packet, therefore, no TinyTP flow control is performed and TinyTP maximum segment size + TinyTP and IrLMP headers equals to IrLAP maximum frame size ( S + llmp + lttp = l ). The length of the packet header is illustrated in Fig. 3(a). The mathematical model uses Fig. 3 for derivation of OBEX throughput using the IrDA protocol stack IrLAP, IrLMP, TinyTP. We make use of Table I for symbol details. 1byte 3bytes 2bytes 1byte 6bytes 4bytes 1byte IrDA Frames BOF IrLAP Overhead IrLMP Overhead TinyTP Overhead OBEX Overhead OBEX Payload CRC EOF (a) OBEX 6byte l OBEX P REQ OBEX Payload: P l REQ OBEX response RES maximum TinyTP segment size receiver TinyTP TTP TTP window Segment TTP TTP size + 1byte 1byte 1byte 1byte 1byte TinyTP overhead + IrLMP 2byte 2byte 2byte 2byte 2byte overhead 9byte IrLAP IrLAP frame IrLAP frame ACK IrLAP frame IrLAP frame remainder ACK 1 byte 2byte IrLAP frame ACK One IrLAP window =2 IrLAP + Physical Layer Overhead: l LAP +l Phy T full +T rem T TA T RES T TA (b) Fig. 3. (a) VFIR IrDA frame structure; (b) Mapping OBEX, TinyTP, IrLMP to IrLAP frames 7

8 TABLE I: IrLAP PARAMETERS USED I THE MODELIG Symbol Parameter Description Unit C Link data rate bits/se p b Link bit error rate - p Frame error rate - umber of frames in one IrLAP window - l I-frame message data length bit l Phy Physical layer overhead: BOF+EOF+CRC bit l LAP S-frame length/ I-frame overhead bit t s Transmission time of a Supervision (S)-frame sec t ta IrLAP minimum turnaround time sec t ack Time to transmit an acknowledgement packet sec t Fout IrLAP F-timer time-out period sec OBEX PARAMETERS USED I MODELIG Symbol Parameter Description Unit S TinyTP maximum segment data length bit P REQ OBEX request packet size bit P RES OBEX response packet size bit l OBEX OBEX request packet overhead 48bit l LMP IrLMP overhead 16bit l TTP TinyTP overhead 8bit T TA OBEX turnaround time sec 3.1. IrDA Stacks Modelling As described in section 2, fixed headers of 2 bytes and 1 byte for IrLMP and TinyTP are considered respectively for each TinyTP segment, 3 bytes and 6 bytes of IrLAP and IrPhy headers for each IrLAP frame, as shown in Fig. 3(a). From the assumption, we have S+ llmp + lttp = l which gives us 12 bytes header in total for each IrLAP frame. The parameters t s, t I, t ack, p and t Fout are defined as follows: t s lphy + llap + llmp + lttp =, C t = t + t according to the IrLAP standard [10]. Fout I 2 ta t I l+ lphy + llap l+ lphy + llap =, t ack = 2 tta + ts, p = 1 (1 p b ) and C From Fig. 1, window transmission time t w denotes the average time needed for a complete window transmission and is given by: t = t + p( t + t ) + t (1) w I Fout s ack 8

9 Since correct frame transmissions following an erroneous frame transmission in the same window are considered out of sequence and have to be retransmitted (GB), the probability p c (w) of successive w correct frame transmission followed by an error is given by: w p ( w) = (1 p p, w =1, 2,, -1 (2) c ) The probability that all frames in a window are correctly transmitted is: p ( ) ) c = (1 p (3) Finally the average number of frames correctly transmitted in one window, cor is: cor 1 = wpc ( w) = w(1 p) w= 1 w= 1 w p + (1 p) (1 p)(1 (1 p) = p ) (4) 3.2. Derivation of OBEX Throughput Since OBEX uses stop and wait as its transmission scheme, the transmitter has to wait for the acknowledgement before sending the next packet. For each OBEX request packet, TinyTP has to transmit a packet with length of P REQ, as illustrated in Fig. 3(b). To send one OBEX packet, it needs several full IrLAP frames and is likely that there will also be a single incomplete IrLAP frame at the end, Fig. 3(b). When l < P, then one OBEX packet requires more than one REQ IrLAP window for its transmission. If l P, the OBEX packet can be accommodated within REQ a single IrLAP window. The transmitter will simply set the P bit at the end of the IrLAP window indicating an acknowledgement request for the receiver. Thus, l P is the only case that needs considering. To transmit a large OBEX packet, several complete IrLAP windows and probably one incomplete IrLAP window are needed. Using equation (4), the average time required for transmitting all the complete IrLAP windows for one OBEX request packet is given in (5): PREQ / S OBEX T = t t (1 p) (1 (1 p) ) / p = cor full w w Where OBEX is the number of IrLAP windows needed to transmit one OBEX packet. The length of the incomplete IrLAP window is given by: REQ L = ( P )mod( S ) (6) rem REQ cor (5) 9

10 The number of frames in the incomplete IrLAP window is: = L / S (7) in rem The probability of having error/errors in the incomplete IrLAP window is: pin1 = 1 (1 p) in (8) Because of the small value of p, p in1 can be approximated as: in pin 1 = 1 (1 p) 1 (1 in p) = in p (9) While error/errors occur in transmitting the incomplete IrLAP window with probability p in1, due to the randomness of error occurrence, it is sufficient to assume that on average the error occurs in the middle of the window, and a retransmission will occur to recover the error with window length of 0.5 in. Furthermore, if further error(s) occur during retransmission with probability of p 0.5in 2 2 in2 = pin1( 1 (1 p) ) 0. 5 in p, retransmission window is required with window length of half the previous value, i.e in, and so on. When the length of retransmission window is less than 1, the whole window is considered to be successfully transmitted. The average time for transmitting the incomplete window therefore can be derived as: 1 1 Trem = inti + p( tfout + ts ) + tack + pin1( inti + p( tfout + ts ) + tack ) + + pin ( inti + p( tfout + ts ) + tack ) 2 2X 1 X( X+ 1) X X = (1 + in p+ + in p ) inti + (1 + pin1 + + pinx )( p( tfout + ts ) + tack ) X ii ( + 1) X ii ( 1) = 1+ i i 2 ( in p) inti in p+ ( in p) ( p( tfout + ts ) + tack ) i= 1 i= 2 2 (10) Where X is an integer with value of X = 0.5 in which satisfies 1 1 X. 2 in Since the response packets are used only for the acknowledgement purpose (no payload), the packet length is equal to OBEX header l OBEX (P REQ =l OBEX ). Due to the small size of l OBEX, it can be accommodated in a single IrLAP frame and we assume it is error free. The time required to transmit a response packet is: T P + l + l + l + l l + l + l + l + l RES TTP LMP LAP Phy OBEX TTP LMP LAP Phy RES = = (11) C C 10

11 Because of the small size of acknowledgement frame and fast hardware link turnaround (IrLAP turnaround), OBEX turnaround time T TA is assumed to be much longer than the IrLAP acknowledgement time, Fig. 3(b). Thus, the average time to transmit one OBEX packet is: T = T + T + T + 2T (12) full The OBEX throughput, which is defined as the useful data bit per second, is given by: D P rem l RES TA REQ OBEX = (13) The throughput efficiency is TPE = D / C (14) T Using equation (14), we compare the throughput for different OBEX packet sizes and IrLAP parameters for a range of BER from 10-4 to 10-8, IrDA data rate of 16Mbps. The OBEX and IrLAP minimum turnaround time T TA and t ta are set to 0.2 ms and 0.1 ms, respectively. Unless otherwise specified these values will be used throughout the paper. The throughput efficiency decreases as the BER increases, see Fig. 4. By comparing the three curves, Curve 1 displays lower throughput at low BER and higher throughput at high BER compared with Curves 2 and 3. The results indicate that different combination of OBEX packet size P REQ, and IrLAP parameters and l have different effect on the system throughput. Throughput Efficiency OBEX Throughput Curve 1 Curve 2 Curve E E E E E-04 Bit Error Rate Fig. 4. OBEX throughput efficiency versus BER using non-optimum parameters Curve 1: = 30, l = 1024bit, P = 512Kbit; Curve 2: = 60, l = 6Kbit, P = 256Kbit; Curve 3: = 80, l = 12Kbit, P = 128Kbit; 11

12 : IrLAP window size, l: IrLAP frame size, P: OBEX packet size 4. PARAMETERS OPTIMISATIO Maximum OBEX performance can be achieved by choosing the appropriate OBEX and IrLAP parameters for a given BER. In this section equations used to optimize the parameters are derived and a number of results are presented to verify the mathematical analysis Optimum OBEX packet size Maximum OBEX packet size P REQ is one of the negotiable parameters for the connection and its value can be chosen from 2Kbit to 512Kbit as defined in [3], it is very important to have a better understanding of its effect on OBEX throughput. By assuming the time to transmit an IrLAP window is equally distributed for each IrLAP frame, the following equation holds true: PREQ PREQ tw floor tw + Trem = (15) S (1 p) (1 (1 p) ) / p S (1 p) (1 (1 p) ) / p P REQ is very large compared to the overhead l OBEX, hence, we can assume P Applying these assumptions to equation (13) it becomes: REQ l P. PREQ lobex PREQ D = PREQ PREQ tw floor 2 RES 2 tw TRES Trem T T T TA S (1 p) (1 (1 p) ) / p S (1 p) (1 (1 p) ) / p (16) Because of small value of p, 1 (1 p) 1 (1 p) = p. By setting OBEX REQ TA A= ( T + 2 T ) S (1 p), (16) is further simplified to: RES TA PREQ S (1 p) D = P t + A REQ w (17) In (17), A is independent of P REQ, and since t w << 1 then tw << S (1 p). Thus D is proportional to P REQ. Because OBEX resides on top of the connection-oriented TinyTP, all the OBEX packets will be transmitted error free. Therefore, by making the assumption in (16), the system throughput will always benefit from a bigger P REQ for any bit error rate. 12

13 4.2 Optimum IrLAP window or frame size for maximum OBEX throughput Besides P REQ, IrLAP parameters and l are the other two major factors for the system throughput. Due to the half duplex nature of the IrLAP protocol, window size is an important and easily adjustable parameter. If a large window size is implemented, the possibility of sending frames following an erroneous frame until the end of the window is higher. All the frames that are received after an erroneous frame are considered out of sequence and discarded by the receiver. These frames essentially delay the reversing of link direction. Time taken for such frame transmissions reduces throughput. In contrast if the window size is small, the link transmission direction changes more often and acknowledgements also have to be sent more frequently. However, it is possible to optimize for different bit error rate p b. By calculating the derivative of D for equation (17) = 0, the optimum value of for any fixed l and P REQ is derived and given in (18). Since the overhead of the physical and IrLAP layer is very small compared to data size, l+ l + l Phy LAP two approximations have been made in the calculation: p = 1 (1 pb) lpb and t l I. C opt 2t ack C = round 2 l p (18) b Besides optimizing window size, frame size can also be optimized to improve the OBEX throughput. A smaller frame size reduces frame error probability and the necessity for retransmissions. However, since each frame transmission requires the transmission of flags, address field, control field and FCS, implementing smaller frame sizes results in relative increase D of overheads. An optimum frame size l can be derived for calculating = 0, for fixed P REQ and l. After some calculations, the following equation for optimum l is derived: l opt tackcpb + 2tackCp b = round p b (19) In order to examine the accuracy of equations (18) and (19), we compare the analytical optimum values with the exact optimum results. The exact optimum results are obtained using a numerical algorithm which locates maximum throughput (using (13)) by numerically cycling the 13

14 integer values of or l in the range of and 512bit-16Kbit respectively for different bit error rates. In Fig.5 and 6, the optimum values of window or frame size l using exact and analytical approaches are compared. The corresponding throughput efficiency for and l values is also plotted in Fig.5 and 6. The results are plotted against the BER by using a P REQ value of 512Kbit. The range of the results is restricted by the IrLAP specification [10]. Thus, optimum is fixed to 127 if a value larger than 127 is required, while the optimum l is fixed at 16Kbit if the results require a value larger than 16Kbit. Optimum opt opt 1.0E E E E E-04 BER opt for l=16k (Eq.18) opt for l=1k (Eq.18) TPE for l=16k (Eq.18) TPE for l=1k (Eq.18) opt for l=16k (Exact) opt for l=1k (Exact) TPE for l=16k (Exact) TPE for l=1k (Exact) Fig.5. Comparison of optimum window sizes by using exact algorithm and analytical model for two different frame sizes of 16Kbit and 1Kbit; The corresponding OBEX throughput efficiency (TPE) is calculated using OBEX packet size P REQ =512Kbit. TPE 0 OBEX Throughput Efficiency 14

15 Optimum l (bits) l opt 1.0E E E E E-04 opt l for =5 (Eq.19) opt l for =50 (Eq.19) TPE for =5 (Eq.19) TPE for =50 (Eq.19) BER opt l for =5 (Exact) opt l for =50 (Exact) TPE for =5 (Exact) TPE for =50 (Exact) Fig.6. Comparison between optimum frame sizes by using exact algorithm and analytical model for two different window sizes of 5 and 50; The corresponding OBEX throughput efficiency (TPE) using OBEX packet size P REQ =512Kbit is also shown. In Fig.5, the optimum results from exact and analytical approaches (equation (18)) are compared. Two different frame sizes are used: l=1kbit, and l=16kbit. As shown in Fig.5, the optimum decreases rapidly when the BER increases. By comparing the exact and analytical optimum values, the major differences are in the low BER. For the high BER, these two approaches match very well. The differences between the exact and analytical results are mainly due to the approximation made in equation (15). The approximation reduces the effect of the incomplete window in the throughput equation. Since the optimum values are large in the low BER, the incomplete window may be large, and hence has a significant effect on the throughput (to be discussed later). However, as the OBEX throughput is already very high for low BER, acquiring accurate optimum values for low BER do not have much improvement on throughput. To obtain accurate optimum values for high BER is much more important since there is more to be gained on the throughput. Despite the difference between exact and analytical approaches for optimum for the low BER, only a small difference is found on the corresponding throughput results, see Fig.5. This shows that the optimum IrLAP window equation (18) is adequate and verifies the validity of the approximations made. l opt TPE OBEX Throughput Efficiency 15

16 In a similar manner, the results of optimum l from exact algorithmic and analytical approach using equation (19) are compared in Fig.6 by fixing at 5 and 50 respectively. All the results have large values at low BER and then decrease rapidly as the BER increases. When =5, equation (19) provides optimum l in an excellent accuracy by comparing to the exact values. When =50, optimum l from (19) is not very close to the exact values especially for the low BER. This is due to the approximation made in (15) (to be discussed later). However, the difference for optimum l between the two approaches at low BER does not affect throughput significantly as shown in Fig. 6. Thus, using equation (19) for optimising IrLAP frame size l is also founded Simultaneous optimisation of IrLAP window and frame sizes for maximum OBEX throughput The best possible OBEX throughput can be achieved when and l are simultaneously optimized with BER. In order to derive optimum and l, the derivative of throughput equation (17) Db Db with respect to and l = = 0 is to be solved. After some calculation, the l simultaneously optimum values of and l are derived: l opt lphy + llap = (20) p b and opt = l 2t ack Phy C + l LAP (21) Equations (20) and (21) reveal that opt is essentially independent of BER, and l opt becomes very large and takes values larger than 16Kbit (the maximum allowed by the IrDA specification [10]) 7 for low bit error rates ( p < from equation (20) by using l + l = 72bit ). In order to b comply with IrDA specification in a similar manner as before, when the results l opt are required to be larger than 16Kbit, l opt is fixed at 16Kbit and use (18) for optimum l. For Phy LAP 7 BER > , however, (20) and (21) can be used for the optimum and l values. It should be noted that the product of l opt and opt can not be larger than OBEX packet size ( l opt opt P send ). If Psend lopt opt > Psend, opt will be set to the value of l opt. 16

17 In order to examine the accuracy of equations (20) and (21), the analytical results are compared with the exact results. The optimum is fixed at 127 if the values result to be greater than 127, while the optimum l is fixed at 16Kbit if it results to be greater than 16Kbit in a similar manner as in section 4.2. Fig.7 shows a comparison between optimum IrLAP window and frame l sizes using exact and analytical. The results are plotted against BER in the range of 10-8 ~10-4. In Fig.7, the curves of exact optimum values for both and l show a non-linear shape throughout the whole BER range. This is due to the changes in the size of incomplete IrLAP window L REM in equation (6). However, in spite of the fluctuations, the analytical results based on our approximations follow the shape of the curves of exact result. The optimum OBEX packet size P REQ and the corresponding throughput efficiency are shown in Fig.8. As also deducted from equation (17), the optimum P REQ should use its maximum allowed value to achieve maximum throughput. Thus, the optimum P REQ uses upper limit of 512Kbit in the analytical approach. In Fig.8, the curve of the exact optimum OBEX packet sizes slightly fluctuates from low to high BER. evertheless, it is very close to the values of 512Kbit which is the optimum P REQ derived from our analysis. This verifies our mathematical analysis for the optimum OBEX packet size. Optimum Frame Size l l opt opt E E E E E-04 BER Optimum Window Size Throughput Efficiency E E E E E-04 BER TPE P REQ 5.0E E E E E E E+05 Optimum OBEX Size Optimum l (Eq. 20) Optimum (Eq. 21) Optimum l (Exact) Optimum (Exact) TPE (Analytic) OBEX (Eq. 17) TPE (Exact) OBEX (Exact) Fig. 7. Optimum IrLAP frame and window size versus BER using analytic and exact derivation Fig. 8. OBEX packet size versus BER using exact and analytical results; The corresponding OBEX throughput efficiency by using the optimum parameters in Fig.7 and 8. 17

18 The corresponding OBEX throughput efficiencies using optimum P REQ, and l are also shown in Fig.8. There are only small differences between the OBEX throughput from the exact optimum and l and those from (20) and (21). These two throughput curves match very well especially at high BER. The OBEX throughput using optimum parameters shows considerable performance improvements by comparing it to the non-optimised cases in Fig.4. Significant improvement is achieved at high BER. For instance, OBEX throughput efficiency of over 0.4 is achieved despite the high BER (BER=10-4 ), while throughput efficiency of 0.1 or less is observed for the nonoptimised cases at the same BER Effect of different protocol tasks on OBEX throughput The relative time consumed on each of the protocol tasks is examined in this section. According to equation (12), the average time T to transmit one OBEX packet consists of the following components: T full, T rem, T RES and T TA. The transmission can be separated into three protocol tasks: full IrLAP window transmission T full, incomplete (last frame) IrLAP window transmission T rem and waiting for acknowledgements from receiver T RES +2T TA. The following equation always holds true T / T + T / T + ( T + 2 T ) / T = 1. full rem RES TA 100% 80% Time Percentage 60% 40% 20% 0% 1.0E E E E E-04 BER full IrLAP windows incomplete IrLAP window acknowledgement time Fig.9. Time percentage against BER for the three different OBEX protocol tasks 18

19 The time percentage of the three different protocol tasks is plotted against BER in Fig.9. Optimum protocol parameters (P REQ, and l) are used. In Fig.9, full window transmission of the main body of OBEX packets is the dominant protocol task for all BER among the three protocol tasks. Incomplete window transmission represents the last part of an OBEX packet which can not fill up one full IrLAP window. It fluctuates in a complementary manner with the full window transmission. Finally the acknowledgement time portion, represents the time spent on waiting for acknowledgements from receiver, is very small by comparison to the other two time portions. The fluctuation of full and incomplete window transmission is due to the changing BER and the corresponding changing optimum and l values. Although incomplete window transmission takes a large time percentage at the low BER, it only consumes a negligible time percentage compared to the full window transmission at the high BER. Therefore, equation (15) holds true in the high BER case. Recall that OBEX throughput gains significant improvement with the optimum parameters only for high BER, therefore the approximation in (15) is sufficient to derive the optimum parameters in 4.2 and EFFECT OF MAXIMUM OBEX PACKET SIZE AD OBEX TURAROUD TIME O THROUGHPUT As discussed in the previous section, OBEX packet size P REQ should be implemented in maximum size to maximise the OBEX throughput. According to [3], the upper limit of 512Kbit will be used for optimum P REQ. Larger P REQ should have better OBEX throughput but larger memory buffer size has to be used to store the large size packet. It is interesting to examine if the throughput can be improved significantly if larger P REQ values were used. We examine the effect of the OBEX packet size P REQ on the link throughput, and search for suitable values of P REQ for the system at different data rates. Using equations (13) and (14), the OBEX throughput efficiency as a function of OBEX packet 4 6 size in the range of 5 10 ~ 1 10 bits, is examined at the BER of 10-6 in Fig.10. In order to minimise the effects of IrLAP parameters, results are produced with optimum and l values. OBEX throughput for data rates of 115.2Kbps, 16Mbps and 100Mbps are plotted in the figure. 19

20 For the three data rates, the corresponding OBEX throughput efficiencies are all increasing with the OBEX packet sizes P REQ. The throughput at 100Mbps benefits most as P REQ increases, while at lower speeds the benefits are small from very large P REQ. For 100Mbps links, if OBEX packet size P REQ is increased to 1Mbit in combination with the use of simultaneously optimum and l values, high throughput efficiency over 0.75 is achieved. P REQ size of 500Kbit and 100Kbit are sufficient for 16Mbps and 115.2Kbps respectively with the use of optimum and l. Larger P REQ is not necessary since it leads to only trivial improvement on the throughput. Throughput Efficiency ) E E E+06 OBEX size (bits) Throughput Efficiency ) E E E E E-01 OBEX Turnaround Time (s) data rate=115.2kbps data rate=100mbps data rate=16mbps data rate=115.2kbps,ber=10e-6 data rate=100mbps,ber=10e-6 data rate=100mbps,ber=10e-4 data rate=16mbps,ber=10e-6 data rate=100mbps,ber=10e-8 Fig. 10. Throughput efficiency versus OBEX packet size at different data rates, at BER=10-6. Fig. 11. Effect of OBEX turnaround time on throughput at different data rates, at BER=10-6 OBEX turnaround time T TA value of 0.2ms is used in all the previous figures. This high layer turnaround time depends on the CPU speed of communication peers rather than IrDA transceivers themselves. It is important to study the effect of OBEX turnaround time on the OBEX throughput and recommend suitable value of T TA for OBEX at various data rates. OBEX throughput as a function of OBEX turnaround T TA time in the range of ~1 10 s is examined at BER=10-6 in Fig.11. In order to highlight the importance of T TA and minimize the effects of other factors, optimum P REQ, and l values are used, and IrLAP turnaround time is fixed at 10-4 s. OBEX throughput efficiency is shown for data rates of 115.2Kbps, 16Mbps and 100Mbps. 20

21 The OBEX throughput efficiency decreases with the increasing turnaround time. The high data rate links are more sensitive and vulnerable to a large T TA than the low data rates. At BER=10-6, very poor OBEX throughput efficiency is observed for 100Mbps when T TA =0.1s, while throughput efficiency of over 0.9 for 115.2Kbps is possible for the same T TA. For the 100Mbps links, an OBEX turnaround time of less than 10-4 s is necessary, while for 16Mbps or less, an OBEX turnaround time of 10-3 s is sufficient if we are not prepared to sacrifice more than a few percent of throughput efficiency. At the low BER (BER=10-8 ) the results are still valid, however, for high BER (BER=10-4 ) the throughput efficiency is significantly reduced and therefore we can afford to be less stringent on OBEX turnaround time the values quoted here. 6. COCLUSIOS OBEX has been widely implemented in the wireless object exchange applications for both IrDA and Bluetooth communications. Although many studies have been carried out to address different issues at the lower layers, there is no systematic analysis for the upper layer. To obtain a better overall performance for a system, it is also important to make sure the upper layer is efficient and compatible with the lower layer. This article examined the performance of OBEX protocol, investigated the interaction between OBEX and the lower IrDA protocol stack, and also optimised different parameters to achieve the maximum system throughput. An analytical model was developed to derive the OBEX throughput equation. Based on the model, the paper examined the impact of OBEX packet size, IrLAP window and frame sizes on OBEX throughput. The system showed very poor performance at high BER with non-optimum combinations of the parameters. In order to maximise the throughout and improve the system performance at high BER, the optimum OBEX size and IrLAP link parameters of the model were studied. The analysis carried out showed the system throughput always benefits by a large OBEX size, and IrLAP frame size l should be reduced gradually when BER is increasing, while IrLAP window size is independent of BER until BER > The optimum equations for and l were given for both individual and simultaneous optimisations and then verified by the exact numerical results. The OBEX throughput showed significant improvements by applying the optimised parameters. Finally, a study of the effects of maximum OBEX packet and OBEX 21

22 turnaround time size was also carried out. Based on the analytical results, we conclude the OBEX parameter selection guidelines of suitable OBEX packet size and turnaround time for different data rates. 22

23 REFERECES [1] Williams S., IrDA: Past, Present and Future. IEEE Personal Communications, 7(1), (2000) [2] Sairam, K.V.S.S.S.S.; Gunasekaran,.; Redd, S.R.; Bluetooth in wireless communication. IEEE Communications Magazine, Volume: 40 Issue: 6, Jun 2002 Page(s): [3] IrDA, Object Exchange Protocol (IrOBEX), Version 1.3 (Infrared Data Association, March, 2003) [4] A. C. Boucouvalas, V.Vitsas, 100 Mb/s IrDA Protocol Performance Evaluation. Proceedings of IASTED International Conference on Wireless and Optical Communications (WOC 2001) June 27-29, 2001 Banff, Canada, pp [5] P. Chatzimisios and A. C. Boucouvalas, IrLAP IrDA Protocol Throughput Dependence on Processor Speed. International Symposium on CSDSP 2002, Staffordshire University, July 15-17, pp [6] A. C. Boucouvalas, V.Vitsas, Optimum Window and Frame Size for IrDA Links. IEE Electronic Letters, Vol.37, o.3, pp , 1st Feb [7] Deccio, C.T.; Ekstrom, J.; Partridge, D.R.; Tew, K.B.; Knutson, C.D.; A study of the suitability of IrOBEX for high-speed exchange of large data objects. Global Telecommunications Conference, GLOBECOM '03. IEEE,Volume: 5, 1-5 Dec. 2003, Pages: [8] IrDA, Serial Infrared Physical Layer Specification Version 1.1, (Infrared Data Association, 1995) [9] IrDA, Serial Infrared Physical Layer Specification for 16Mb/s Addition (VFIR) Errata to version 1.3, Infrared Data Association, (1999) [10] IrDA, Serial Infrared Link Access Protocol (IrLAP), Version 1.1 (Infrared Data Association, 1996) [11] IrDA, Link Management Protocol (IrLMP), Version 1.1 (Infrared Data Association, 1996) [12] IrDA, Tiny Transport Protocol (TinyTP), Version 1.1 (Infrared Data Association, 1996) [13] K. Saramas; Street, A.M.; O'Brien, D.C.; Edwards, D.J. Method for evaluating error rates in infrared wireless links. IEE Electronics Letters, Volume: 33 Issue: 20, 25 Sep 1997 Page(s):