SPECIFICA‘TION, VERIFICATION AND SIMULATION OF A WIRELESS LAN                                PROTOCOL: MACAW              ...
backoff algorithm oscillated too rapidly. Instead the          predicate. The full specification is given in [All.value of...
environment greatly eiased the simulation process.                              These cases are also tested from 10% to 10...
maximum performance of the node that is running                network increases, since per node utilization and dataMACAW...
arrival rate value). The “recognition” period in the               observed, which was caused by the rapid iiocrease of th...
Upcoming SlideShare
Loading in …5
×

Specifica‘tion, verification and simulation of a wireless lan

278 views
221 views

Published on

Published in: Education, Business, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
278
On SlideShare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
0
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide

Specifica‘tion, verification and simulation of a wireless lan

  1. 1. SPECIFICA‘TION, VERIFICATION AND SIMULATION OF A WIRELESS LAN PROTOCOL: MACAW G M Lundy, M. Almquist, T. Oruk U S Naval Postgraduate School Monterey, CA 93943 ABSTRACT able to make some suggestions as to how it can beA formal specification and verification of a wireless improved, and to give a fairly accurate indication of howLAN protocol is given, and a simulation of the much the improvements will help performance.protocol’s performance is also described. The verification is a partial verification in that it does not MACAW -- Multiple Access, Collision Avoidance, include a general verification of all possible Wireless -- was designed as an improvement upon an configurations of the protocol; however; it does cover earlier protocol, called MACA [Karl. MACAW is based the expected usage of ihe protocol, which is for a small on the same principle as is the cellular radio network for number of users sharing a cell. The simulation provides automobiles. A set of base stations, which are ailfurther insight into the protocol, especially concerning connected by a wired system, communicate via radio its performance. Ba,sea! on the verification and signals with the user stations. The difference is that withsimulation, some sug,gestions are made which can MACAW, the base stations are connected by a local area improve the protocol’s performance. network -- such as CSMACD -- rather than a telephone type network covering a large area. The area aroundI. INTRODUCTION each base station is called a cell; and the usersWireless communications in various forms has been the communicate with the nearest base station. Anothersubject of much attention and research in recent years. difference is in the cell size; here we are using cellsWireless communications include the broadcast radio which are only a few meters in diameter, rather thanmedium; terrestrial microwave[Mili]; satellite microwave miles across. A third difference is that the :stations arein all its forms [Ha,Ree]; and the International Amateur not mobile, as are autos. The signals are relatively weak,Radio Network, or “Ham Radio”[Bi]. Additionally , so that they should not penetrate beyond the “cell,” orwireless communications includes acoustical (sound) its local area. However it is possible that somecommunications, used by submarines for many years, interference between adjacent cells may occur. Otherand also for data communications, at least protocols for wireless and broadcast networks exist, andexperimentally [Rei]; and optical (visible) light have been in use for years; these are discussed in thecommunications, which are still used by maritime references.vessels. More recent fioms of wireless communicationsincludes wireless telephones; cellular telephones; air MACA (medium accessicollision avoidance) Karl,phones, which enable the passenger aboard an airliner to MAC‘4W’s predecessor, used a sequence of three basicconnect to the long distance telephone system; and messages: request-to-send, clear-to-send, and DATA.wireless computer networks. Pagers and remote control MACX used binary exponential backoff aligorithm todevices might also be included, though these are very resolve collisions, the same type used in CSMA/CIl.“low bandwidth” communications devices. Studies showed that the MACA protocol was not fair to all stations on the network, and the throughput was notIn this paper we discuss a protocol suggested for wireless very high under conditions of imperfect channels [BD].!oca1 computer networks, known as MACAW [BD]. Thisprotocol was suggested as an improved means to enable MACAW (MACA-wireless) improved upon IMACA bycomputers in a local area to communicate with each adding to this message sequence. The request-to-sendother. The protocol is intended for a single channel (KTS) message from the sending station is followed bypacket radio network. This means that all stations use the the clear-to-send (CTS) from the receiver, as withsame channel (frequency), so that should more than one MACA; but the DATA is preceded by a short data-sendtransmit at a time the signals may interfere. (DS) packet, which serves to alert the nearby stations that a data packet is about to be transmitted, reducing theNext we briefly describe MACAW. In section 2 , the probability of a collision. This is followed by an ACKformal specification is summarized. Section 3 discusses packet from the receiver, through which reliability isaverification of the protocol.Further study of the enhanced. This means that MACAW is able to provideprotocol was carried out through a series of simulations; reliability at the MAC layer, where MACA co’uld not.this is described in section 4. As a result of this work -what we have learned through the formal specification MACAW also uses a different backoff algorithm thanprocess, the verification, and the simulation -- we are MACA. MACAW’S developers felt that the MACA U.S. Government work not protected 565 by U S . copyright.
  2. 2. backoff algorithm oscillated too rapidly. Instead the predicate. The full specification is given in [All.value of the backoff counter is increased by a factor of1.5 (rather than 3) for each collision, and decreased by 111. VERIFICATION OF MACAW1 for each success. This provided a gentler fluctuation A major benefit of formal specification of protocols orin the backoff counter. computer programs is the potential for automated generation and analysis. This automated analysis wasIn this paper we show that MACAW is a reasonable conducted using the program developed in [Bu]. Theprotocol for its intended purpose, however that it can be analysis is done by constructing a directed graph ofslightly refined and impioved. Our contributions states and the transitions between states.include the formalizing of the protocol specification, itsverification, additional knowledge of its performance, Three basic configurations were analyzed with theand suggested improvements. automated analysis. The number of machines was vaned by using three input files with two, three and four ECIFIC~~IO~ OF MACAW machines. Though it is possible to have more than fourMACAW is formally specified and analyzed using a machines in a single cell, this is unusual; a cell is only amodel called systems of communicating machines. This few meters in diameter. So, while our verification doesmodel, originally defined by Lundy and Miller[LM] has not cover all cases, it does cover the majority ofbeen used to analyze several communications protocols. practical cases.Each machine is defined as a finite state machine withvariables. Communication between machines is done All of the the possible specified transitions of MACAWthrough the use of shared variables. Local variables are were not always executed by all machines; in fact, forused to maintain the state information for each machine, some cases, all transitions were not executed by anyand a predicate-action table serves to unify the state machines. At first glance this appeared as a protocolmachines and the variables. The goal of formal error -- an unexecutable transition but in fact was not.specification is to determine if the protocol is free from It was simply a case in which the transition in questionerrors, such as deadlock or unexecutable transitions, was not needed. For example, in the two machineand if it possesses “liveness,” which is defined as the analysis, none of the quiet stare transitions wereability to make positive progress. An additional benefit executed for the simple reason that there was no otheris identification of ambiguities in the protocol. machine causing interference.A formal specification of MACAW was developed in With two machines: Machine 1 and machine 2 were run[All, which includes a set of local variables, a set of in the two machine analysis. Both began in the IDLEshared variables, used for communication, a finite state state and machine 1 had a message to send to machnemachine, and a predicate-action table. 2. Once machne 3 received the DATA message from machine 1 it sent a DATA message to machinel.The shared variables are channel.type, channel.DA, esn, (Machine 3’s action upon receiving a DATA messagechannel-my-backoff, channel-local-backofS, and included writing a DATA type message to its own outputchannel-retry-amount. These are used for buffer). No deadlock occurred and the total number ofcommunication between the nodes, and the names are system state generated was 69. The correspondingdescriptive of their uses global state analysis generated 275 states. One artificial deadlock occurred in the global state analysis uponThe 1oca1 variables are inbuffer, outbuffer, backof, receipt by machine 2 of the ACK message frommy-backoff, local-backoff. Again, the names of the machinel. The deadlock happened as a result ofvariables are descriptive of their uses. clearing the output buffers when an ACK message was received. Essentially, both machines received theirThe finite state machine diagram has 10 major states. respective DATA messages and no more data remainedThese states are named IDLE, CONTEND, XMIT, to be transferred, thus causing a deadlock.WFCTS, WFACK, WFDS, Q U E , WFDATA, FCONTEND, CQNTEND2. Simulabons with 3 and 4 machines, which generated many more states, are reported in [All.The names of the states are descnptive; in IDLE, thenode has nothing to transmit. In CONTEND, the station IV. SIMULATION AND ANALYSIS OF MACAWhas a message to transmit and is trying to get the The purpose of the simulation is to determine thechannel; XMIT, is sending; “WF” signifies “Waiting behavior of the protocol under different operationalfor,” as in WFACK, “waiting for acknowledgment.” conditions. Therefore, operational conditions are firstThe initials CJTS signify “clear to send.” determined as test cases and simulations are performed with these test cases. The results are analyzed and someThe predicate-action table unifies the finite-state suggestions to improve the protocol are given.diagram with the local and shared variables. Eachtransition in the diagram has an entry in the table, and The simulation model of the protocol was built usingfor each entry there is an action and an enabling OPNET 2 . k from M I D , Inc. The OPNET design 566
  3. 3. environment greatly eiased the simulation process. These cases are also tested from 10% to 100% of the total offered overlapping channel loads. AnA.Test Cases. The behaviors that we are interested in overlapping channel is the intersection of transmissionare the utilization of the channel and packet mean ranges of two (or more) non-communicating nodesqueue delay time. The usual performance measure for a where one’s transmission interferes with the other. .Aprotocol is network utilization, but this does nut consider brief explanation of each case follows:users of the network. ‘Therefore, mean queue delay timeis also considered as a performance measure of the Hidden node: While A is sending to B, C attempts toprotocol. These measures are related to the send to B also. In this scenario A and C are not aware ofCommunication channel load. The load of the channel each other but affected from each other’s transmission,depends on the number of the nodes that are sharing the so they are hidden from each other. &)osed node:same transmission media (channel) and their (internal) While B is sending to C, , attempts to send 1.0 B. Here B 4packet arrival rates. Besides these, the performance of is aware of A, thus B is exposed to -4’s transmissions.the radio networks is greatly affected by hidden- Hidden and exposed nodes: While A and B are mutuallyexposed nodeicell conditions [Karl. exchangmg data, C attempts to send to R. Here A and C are hidden from each other and B is exposed to both AWith these consideratiions in mind, the test cases are and C. Hidden Cell: B and C are in different cells butseparated into two groups known as “load cases” and they can hear each other. Node . is not aware of the 4“hidden-exposed nodei’cell cases.” In the load cases, it communication going on in the adjacent cell since itis also determined whether the protocol acts in favor of cannot hear anything from that cell, it is hid’den from A,any specific communication topology. For each test the same is true for node D. Exmsed Cell B and C arecase, communication scenarios are determined and, for in different cells but each can hear the other’seach scenario, test cases are generated with different transmissions while they are communicating with othernumbers of nodes and packet arrival rates. nodes in their cells (A and D respectively). So they are exposed to each other’s transmissions. ]Hidden andWith each load case, networks of up to four transmitting Exposed Cells: Node A does not hear any tiansmissionsnodes are built and each of these cases is tested from from the other cell xwhle node C is hearing node B. So10% to 100% of the total offered channel loads. Having the adjacent cell is, hidden for A and exposed for C. Allup to four transmitting nodes for each load case in one: Heavy communication in both cells.provided enough information about the behavior of .theprotocol under these conditions. Sample load case B. Performance Measure Criteria. As mentionedscenarios are shown in Figure 1. The load case above, the performance measure also includes the meantopologies of interest are as follows: delay time in addition to utilization. The resultant formula is:{a) reporting nodes with two, three and four transmittingnodes; (b)one-wav communicating nodes with two, Total-Performance = utilization I mean-queue-delaythree and four transmitting nodes; (c)=communicating nodes with one and two pairs mutually According to this formulation, if two utdizations arecommunicating nodes; (d)ring communication with equal then the one with lesser mean delay time will havethree and four nodes. a better performance measure. For example, suppose we have measured an utilization of 54% with a mean delay time of 9 sec and for another case 54% Utilization with a mean delay of 3 sec. Just looking at the utilizations does not say anything about which one is better. But when delay times are considcrcd, the first case will have a performance measure of 0.54/9=0.06 while the second case will have a prformaiuce measure of 0.%/3=0.18. This clearly shows that the second case is better. A performance measure should be independent of transmission speed of the network in order to be used to compare different protocols. Thus, delaiy time isFigure 1: Sample load cases from each communication topology normalized to data packet length. This me.ans we cantested. For all cases above, all nodes are in one cell, that is all can compare the performance of the protocol for networkshear each other and arrows show the direction data streams. (a) four with different transmission speeds.reporting nodes, (b) two 1-way nodes, (c) two pairs cmss-communicating nodes, (d) ring communication with four nodes. C. Test Results. For all cases the Poisson distribution function is used for the packet arrival distnbution. TheHidden-exposed nodeicell cases are shown in [Or]. first test case, our base case, is chosen to determine the 567
  4. 4. maximum performance of the node that is running network increases, since per node utilization and dataMACAW protocol for a given arrival rate. This test is mean queue delay times does not change.conducted with a network of one transmitter and onereceiver, and utilization is calculated. Data mean queue Hidden and Exposed Node/Cell Case Results In thedelay times are collected by OPNE3”s built-in utilities. hidden node case, due to the rapid increase of theOPNEX calculates the delay times of the packets that backoff value of node B, performance is poor abovearrive to the queue during one unit of time and uses its 1 % of overlapping channel loads (load over node B). 0mean (middle) value as a mean delay time of this time Channel access is fair, because both transmitters areperiod. The simulation result of this base case and the using the same backoff value as in reporting nodes caseother cases are shown in detailed tables in [Or]. of load cases.As seen from the table, the performance decreases In the exposed node case performance is fairly highrapidly beyond 80% offered channel loads due the high compared to the hidden node case. However, above 80%mean delay times. Maximum utilization observed for of offered overlapping channel loads, the behavior ofthe protocol with given packet sizes (30B for control the network becomes similar to thc one waypackets, 520B for data packets), packet arrival communicating node cases of the load cases. Thedistribution function and backoff limits is 66%. These “lucky” node gets the channel and keeps it.results will be a base case to use in comparison with testresults. In the hidden and exposed node case , the hidden node behavior is dominant. Due to the high delay times, theLoad Case Results For the load cases, channel performance decreases dramatically above 4%ofutilization almost never exceeds 50% for all simulations overlapping channel loads .(the maximum utilization observed for the protocol was66% with current parameters). The exceptions are the In the hidden cell case channel access of transmitterbase case and “one pair mutually communicating” nodes (node -4 and D) are always fair because thenetwork case. In the base case there is no interfering congestion levels at the receivers are similar. But due tonode against the transmitting node, so we should get the congestion at the receiver side data packet delay“the most” from the protocol. The reason that one pair times in the transmitters are h g h so performance is lowmntually communicating network has high utilization for cell loads above 25%. Its behavior is the same as theand low backoff values than the other simulation cases is hidden node case.it has a slower rate of backoff increase for its remotebackoff values. Modified Protocol Simulations During the simulations it was observed that the backoff increase rate is one of theFor high loads, one-way communicating nodes cases key factors that affects the performance of the MACAWalways violate the fairness of channel access. The protocol under moderate and high loads (40% andbackoff algorithm dictates that each node keeps track above). To further investigate, the same set ofof the congestion of the channel between itself and the simulations were run with a lower backoff increase rate.neighboring nodes via backoff values. When a node As a result, for load cases with a small number ofattempts to communicate with one of its neighboring transmitting nodes, significant performance gains werenodes, it uses the associated backoff value. These observed. These gains are due to the lower mean delaybackoff values are mutually exclusive. In a one way times rather than increased utilization. But as thecommunicating network scenario when a “lucky” node number of the transmitting nodes increased, thereaches the channel quicker than the others for a couple performance gain achieved by this modificationof consecutive times it gets and keeps the channel by degraded. No significant performance change wash o w i n g other nodes have very high backoff values. observed in hidden-exposed nodeicell cases.Reporting and ring cases have similar results. They both Another factor that affects the performance of thehave fair channel access during the simulations. But protocol is the nodes’ inability to determine whether theboth have high backoff values. To have high backoff channel is busy. According to the MACAW protocol,value causes larger delay times on average therefore nodes behave purely in accordance with the receivedreporting and ring cases have lower performance packet information. The keyword here is “received.”measures. The MAC layer that runs the MACAW protocol does not know that it is receiving a packet until the entire packetThe best performance measures observed during the has been transmitted. The probability of collision of twosimulations are around 20% channel load conditions for transmitting nodes is the probability of transmission ofthree and four transmitting nodes cases. A s the number one node without recognizing (or before recognizing)of transmitting nodes in the network increases, the effect that another node is transmitting. The transmissionof having high backoff values decreases (for low loads), probability of a node when it assumes that the channel isbecause when the channel is free, the chance that there is available is based on the distribution function and itsa node ready to transmit is high. So idle times of the parameter value (in this case Poisson distribution withchannel are used and total performance measure of the 568
  5. 5. arrival rate value). The “recognition” period in the observed, which was caused by the rapid iiocrease of theoriginal MACAW protocol is the time that it takes for backoff increase rate for the small number ofcontrol packet transmission plus propagation delay. transmitting nodes cases. 4 relation between the backoffThus, we haw a sample space of increase rate and the number of transmitting nodes wastime_slot+propugatic,n_clelay for a transmission event observed (but this needs to be investigated morethat may cause a collision. When carrier sensing is thoroughly). So, instead of using a linear increase rateintroduced to the protocol, this sample space shrinks to use of dynamic or exponential increase rates may giveone propagarion-delay ( assuming time-slot >> better results. The dynamic increase can bind to thepropagation delay). So, carrier sensing will decrease the number of transmitting nodes. For the fairness problemprobability of c:ollisions by a factor of in one-way communicating nodes cases moretime-slotlpropagatioi~z_delay. This is also what is fundamental changes should be made to the algorithm.observed during the simulations of the MACAW That is, the backoff algorithm needs to be thoroughlyprotocol with carrier sensing capability. Significant analyzed.performance measure: increases observed for SO% andabove common channel loads. The source of gain is It is also observed that the channel acquisition withlower mean delay times. While the greatest performance carrier sensing is superior to the original MACAWgain is achieved in the hidden node case (for SO% protocol policy. This is quite significant for mediumoffered load), no change was observed in the hidden cell and high loads. This contradicts the reasoning of thecase. In hidden cell case carrier sensing provides no help original MACAW protocol specification which saysto the network to improve its performance, because “...carrier sensing is inappropriate” [13D]. Thistransmitters are out of range from each other (the only simulation is discussed in more detail in [Or] , as well astime that a transmitter utilizes carrier sensing in this the remainder of the simulations which are omitted heretopology is to sense its receiver’s RRTS packet). due to lack of space..4nother improvement achieved by carrier sensing is V. CONCLUSIONSfairer channel access of the transmitting nodes Uk described the: specification, verification, and(compared to the original network) in the “hidden- simulation of MACAW. The specification uncoveredexposed cell” and “aill-in-one” network topologies. some ambiguities in the original specilfication. The verification showed the protocol to be free of deadlocks,In “hidden-exposed cell” network node C cannot get for most practical cases. Simulations gave furtherthe common channel1 as frequently as in the origmal confidence in the protocol and showed further possibleprotocol. Now C should wait B to send a11 its packet and improvements in performance. We showed that theact accordingly (nod,e C must understand what B is backoff algorithm is of major importance to the“saying”). ’Ihis increases the performance of node A performance of the protocol, and topology is alsoslightly, because backoff values are dominant for the critical. The protocol can also be improvled by addingperformance and here there is a big difference between carrier sensing. Simulations suggest that this couldcongestion levels of the receivers (especially for high improve utilization under heavy load by about 5%.loads). REFERENCESIn “all-in-one” case fairness is achieved between inner [All Almquist, Marty, Formal Spec$cation and Analysis ~7f a Wireless Media Access Protocol, MS Thesis, Department ofnodes. In the originail protocol for high loads (go%), Computer Science, Naval Postgraduate Schcol, Monterey, CA,one of the inner noldes has much higher performance 1995.measure than the other. Now with carrier sensing both [BDl Bharghavan, V., Demer;, A.. Sbenker. S. and Zhang, L. “MACAW A Media Access Protocol for Wireless LAN’s,”have similar performances. The performance difference Proceedings of ACM SIGCOMM 94, pp. 212-25, ACM, 1994.between inner and outer nodes are still great because of [BUI Bulbul, B. AProtocol Validator,for the SCM and CFSMModels, MS Thesk, Naval Postgraduate School, 1993.the congestion levels at the receivers. CEPl Epstein, M., et al, “Application of Commercial Wireless LAN Technology to Forward Area Mobile Communication,“Simulation Summary. It was determined that the Proceedings, Military Communications Conference, (MILCOM), 19%.backoff algorithm has a significant importance for the [Gel Geier, Jim, Wireless Networking Handbook, New Riders, 1996.Performance of the protocol, where performance is l I E l l- - l IEEE P802.11, Drafl Standard for Wirei’ess LAN Mediumdefined in terms of channel utilization and transmitter Access Contrvl (MAC) and Physical Luyer Specifiraation (PHY), 19 J d y 1996.mean delay times. [Karl Karn P., “ M C A - A New Channel Access Method for Packet Radio,” ARRLICRRL, Amateur Radio Yth Computer Networkmg Conference, September 22, 1990.Backoff is the measure of the congestion at the location [Kat] Katz, R, “Adaptation and Mobility in Wireless Informationof the node. Distribution of this information throughout System,”IEEE Personal Communications, First Quarter 1994.the network provides fair access to this node by other &MI Lundy, G., and Miller, R, “Specification ancl analysis of a data transfer protocol using systems of communicating machines,”nodes since each node keeps track of the other node’s Distributed Computing, May 1991.congestion information in the network. But this does not D l Oruk, Tufon. Simulation and Analysis o a Wireless MAC f Protocol: MACAW, MS Thesis, Department of Computerhelp in the situations like one way communicating nodes Science, Naval Postgraduate Schml, 1996.cases. Also some performance degradation was 569

×