Synchronization and Beaconing in IEEE 80.s Mesh etworks Alexander Safonov and Andrey Lyakhov Institute for Information Transmission Problems E-mails: {safa, lyakhov}@iitp.ru Stanislav Sharov Moscow Institute for Physics and Technology E-mail: sharov.stanislav@gmail.com Abstract Synchronization and beaconing attract much attention with the appearance of IEEE 80.s Amendment draft specifying mesh networking based on IEEE 80. concepts. This paper discusses the drawbacks of using unsynchronized beaconing and focuses on beaconing of synchronized mesh points which adopt beaconing algorithm for IBSS networks with some modifications. We provide an analytical model of the beaconing process and prove its accuracy with simulation results.. Introduction The original IEEE 80. standard [] supposes that one of the main purposes of beacon frames is exchanging timing information between stations (STAs). In an infrastructure network, the access point is the only STA responsible for broadcasting beacons. In an ad-hoc network, all STAs transmit beacons. Anyway, beacons carry information which is used by Timing Synchronization Function (TSF) [] maintaining STAs clock synchronization. By means of TSF the global, or in other words network-wide, synchronization is provided, i.e. all STAs use common time reference points. The global synchronization and beaconing concepts, which remained unchanged for a decade of IEEE 80. standard life, attract much attention with the appearance of IEEE 80.s Amendment draft specifying mesh networking. Compared with legacy IEEE 80., a mesh network provides additional services, and beacons are responsible to support them. In the first mesh networks standard draft [3], mesh- STAs, called Mesh Points (MPs), may choose either to support global synchronization or not. Unsynchronized MPs send beacons as in infrastructure networks, while synchronized MPs adopt ad hoc networks beaconing with some modifications. In the current IEEE 80.s interim draft [4], another step is made towards unsynchronized mesh networking. Per-neighbor based synchronization is used instead of global synchronization. MPs tend to send beacons as access points which radio ranges overlap, taking no steps to prevent their beacons from collisions with data transmissions from hidden MPs. It is impossible to ignore the presence of hidden nodes in a mesh network, so beaconing without global synchronization in a mesh network is very much unreliable. If no common time is maintained, it is difficult to educate hidden nodes to respect beacons from one another and suspend data transmissions for the time of beacon transmissions. With the appearance of mesh networks, it is clear that beaconing requires much attention and accurate analysis. Synchronized MPs may support mesh-wide QoS and efficient power saving which are widely believed to be absolutely necessary for modern Consumer Electronics devices. In this paper, we study the beaconing of synchronized MPs in a network operating according to the mesh standard draft [3]. The rest of the paper is organized as follows. In the next sectio we summarize the evolution of beaconing in IEEE 80. networks and overview the investigation in this field. Section III provides analytical model of beaconing process in mesh networks [3]. We show numerical results obtained with the analytical model and validate their accuracy with simulation results in section IV. Finally, section V concludes the paper.
. Overview of beaconing in IEEE 80. This section summarizes evolution and investigation of synchronization and beaconing mechanisms in IEEE 80. networks, starting from the original IEEE 80. standard [] published in 999. Special attention is paid to the first draft of IEEE 80.s Amendment [3] specifying the operation of mesh networks based on the concepts of recently published IEEE 80.-007 []. In the original IEEE 80. standard [], beacon frames are used to provide STAs clock synchronization and advertise STAs capabilities. Clock synchronization is important for both PHY and MAC layers, e.g. for frequency hopping spread spectrum technology to ensure that all STAs hop at the same time, or for performing power management. IEEE 80. standard [] supports infrastructure and ad hoc modes, called BSS and IBSS, respectively. In a BSS network, beacons are only sent by the access point. In an IBSS network, beacon generation is distributed, i.e. all members of the IBSS participate in beacon generation. The STA which instantiates the IBSS defines a series of Target Beacon Transmission Times (TBTTs) exactly Beacon Interval time units apart. At each TBTT, every STA suspends decrementing its backoff counter for any non-beacon and non-atim traffic and schedules a beacon transmission by setting a timer to a randomly chosen number of slots uniformly distributed between zero and acwmin. STAs decrement their timers using the same algorithm as for backoff. If STAs sense the medium idle during a slot they decrement their timers by. If a STA starts a beacon transmissio other STAs freeze their timers until the medium is sensed idle for DIFS. If a collision happe timers are frozen until the medium is sensed idle for EIFS. If the timer expires, the STA starts its beacon transmission. If a beacon arrives before the timer expiratio the STA cancels its own pending beacon transmissions. At each TBTT, an ATIM Window starts where beacons and ATIM frames may only be sent. To decrease the probability of collisio other transmissions are prohibited within ATIM Window. ATIM frames are used for power save management and are out of scope of this paper. We assume that no STA is in power save mode. Upon receiving a beaco a STA adjusts its TSF timer based on the timestamp indicated in the beaco synchronizing its clock to the beacon. Some TSF problems are well-known. So, TSF is not scalable, as shown in [5]. The beacon collision probability increases with the number of STAs in IBSS. If no beacon is transmitted successfully several consecutive Beacon Intervals STAs may get out of synchronization. Another problem is that the TSF timer can be adjusted forward only, but not backward. More specifically, upon receiving a beaco a STA adjusts its TSF timer based on the beacon timestamp if the value of the timestamp is later than the STA s TSF timer. As a result, fast STAs cannot adjust their timers even if a beacon generated by a slow STA is received. The effect is called the Fastest-station asynchronism and studied in depth [6]. With the appearance of the first completed mesh networks standard draft 80.s/D.00 [3], the synchronization and beaconing rules evolved. An MP may choose to be either synchronized or unsynchronized. Unsynchronized MPs transmit their beacons as access points in BSS networks, each maintaining independent TSF timer and TBTTs. In contrast, synchronized MPs attempt to maintain a common TSF time called the Mesh TSF time and schedule their beacon transmissions at the same TBTTs. In [3], Synchronized MPs adopt IBSS beaconing rules, except for the following aspect. If an MP s beacon is received in a Beacon Interval, other MPs may cancel their own pending beacon transmissions, but they do not have to. It may be inappropriate for an MP to cancel its own beacon transmission. A mesh network may need more than one beacon appeared in a Beacon Interval from an arbitrary MP. Compared with an IBSS, a mesh network provides additional services and beacons are responsible to support them. So, Mesh Deterministic Access (MDA) [3] is a channel reservation method designed for isochronous delay-sensitive traffic. Potentially it is the best method to ensure mesh-wide QoS for multimedia applications. Both MDA transmitter and receiver maintain their MDA reservations by including MDAOP Advertisements information element into beacons, otherwise MDA reservation lifetime expires. This additional information makes beacons from two MPs sufficiently different, compared with beacons in the legacy IBSS, where the only timestamp field value is updated for each STA, while other fields represent STAs capabilities and are not changed during the IBSS existence. So, each MP may need to send its beacon more frequently than STAs in IBSS. In addition to BSS and IBSS beaconing, 80.s/D.00 draft introduces the Beacon Broadcaster (BB) concept. For power saving purpose, MPs may choose one MP to serve as the BB for some time.
When the BB is elected, other MPs generate no beacons. MPs hand over the responsibility for serving as the BB periodically. However, in a mesh network some MPs are hidden from each other, so multiple BBs may appear and the BB rotation problem appeared very complicated. In later versions of the mesh network draft, e.g. in [4], the usage of BB is excluded. In the current interim version of the 80.s draft [4], MPs synchronization is per-neighbor based. o mesh-wide synchronization is provided. With independent TSF timers and TBTTs, the BSS and IBSS beaconing seems to have little difference. In both cases, MPs send their beacons independently, without any coordination with neighbor MPs, except the attempts to escape repeated beacon collisions. The work on the 80.s draft is not finished yet. It is not clear which beaconing rules will be finally adopted as mesh beaconing. We believe that meshwide synchronization and IBSS beaconing make available such important features as mesh-wide QoS and efficient power saving which are core Consumer Electronics requirements. In contrast, in a network without global synchronizatio it seems to be more difficult, if not impossible, to provide QoS. So, in this paper, we focus on beaconing of synchronized MPs in the only completed version of IEEE 80.s mesh draft [3]. 3. Analytical study In this sectio we show the analytical study of synchronized MPs beaconing in an IEEE 80.s mesh network [3]. To estimate the beaconing performance, we calculate the average number of beacons transmitted successfully per a Beacon Interval and the probability for an MP to transmit its beacon in a Beacon Interval. Transmission range of a STA depends on the transmission rate, and it is wider at low rate. Beacons are always sent at the lowest basic rate, while data may be sent at much higher rates. For OFDM technology which is considered in this paper, beacons are sent at 6Mbps, while data may be sent at up to 54Mbps. So, even if MPs are within one another s transmission range at the lowest basic rate, in order to exchange data, the MPs may have to transmit over several hops. Let us consider a network consisting of MPs. To simplify the analysis we assume that every MP can receive beacons sent by all other MPs. Pending a beacon transmission in the beginning of a Beacon Interval, every MP sets a timer to a random number of slots chosen from the interval of K slots. STAs decrement their timers using the same algorithm as for backoff described in section II. The slot equal to aslottime is used as the basic time unit in our model. We represent the beaconing process as the sequence of virtual slots which starts at each TBTT. If no beacon transmission was started in a virtual slot then its length is equal to the length of slot. If one MP starts a beacon transmission in a virtual slot then its length ts is equal to the time needed to transmit a beacon plus DIFS. If several MPs start beacon transmissions in a virtual slot then a collision happens and its length t с is equal to the time needed to transmit a beacon plus EIFS. Both times, t s and t с, are rounded to the integer number of slots in our model, to simplify the analysis. In the next sectio we show that this approximation is accurate enough by comparing numerical results obtained by analytical and simulation models. To increase the frequency of beacons sent by self, an MP does not cancel its pending beacon transmission even if it receives a beacon sent by another MP, but it tries to send its own beacon unless the end of ATIM Window is reached. All pending beacon transmissions which start out of ATIM Window are cancelled. In the analysis, we calculate the average number of beacons transmitted successfully per a Beacon Interval, B(, K, M), where M is the ATIM Window size. Obviously, B (, K, M ) = for every M K, and B (,, M ) = 0 if > and M. As all MPs compete for beacon transmission using common rules, the probability that an MP transmits its beacon in a Beacon Interval is equal among all MPs and may B (, K, M ) simply be calculated as. We consider beaconing process in ATIM Window, proceeding from virtual slot to virtual slot. At the current virtual slot, n MPs have not transmitted their beacons, k virtual slots remain unconsidered, and m slots remain in ATIM Window. The probability that exactly j of n MPs schedule their beacon transmissions in the current virtual slot is j n j n p( j, k) = C j, k k where n! C n j = is the number of ways to j!( n j)! choose j MPs among MPs. A beacon is sent successfully in the current virtual slot, if and only if an MP is the only one scheduled its beacon transmission in the slot, which happens with the probability p(, k). The probability that no MP
schedules its beacon transmission in the current virtual slot is p(0, k). Following along the sequence of K virtual slots, slot by slot, we calculate B(, K, M) recursively: B( k, m) = p(0, k)( k > & m > ) B( k, m ) + + p(, k) + n j= { + ( m > t & k > ) B( n, k, m t )} s p( j, k) [ ( m > t & k > ) B( n j, k, m t )], where (condition) means a Boolean indicator that the condition within the brackets is held. In the recursio the first of three components describes the case when the current virtual slot is empty. If there is at least one more virtual slot to consider (k >) and at least one more slot in the rest of ATIM Window (m >), we add to the current B( ) the average number of successful beacon transmissions in the rest of ATIM Window, with the probability p(0, k). The second component describes the case when exactly one MP transmits its beacon in the current virtual slot, which happens with the probability p(, k). We add one more successful beacon to the current B( ), check that the end of ATIM Window is not reached ( m > ts ) and at least one virtual slot remains unconsidered (k > ), decrease the number n of MPs by and the ATIM Window remainder by t s, and proceed to the next virtual slot. The last component of the recursion describes the case of beacon collision when exactly j = {,, n} MPs transmit their beacons in the current virtual slot, which happens with the probability p(j, k). We check that the end of ATIM Window is not reached ( m > tс ) and at least one virtual slot remains unconsidered (k > ), decrease the number n of MPs by j and the ATIM Window remainder by t с, and proceed to the next virtual slot. We continue calculating B( ) recursively until the end of ATIM Window is reached or all K virtual slots in the current Beacon Interval are considered. This is the end of our analytical investigation. In the next sectio we show the numerical results of the investigation. 4. umerical results c s c + varying the ATIM Window size, M, and the number of MPs in the network,. For OFDM PHY which we consider in our numerical analysis, values of parameters which are important in our analysis are listed in Table. The beacon frame length is the same for all MPs. We calculate the minimal reasonable beacon length, including into the beacon all mandatory fields, but no optional fields, as in [3]. In the scenario we consider, one of MPs in the network maintains a Broadcast MDA reservation [4, 7] for a video stream by including an MDAOP Advertisements informational element into its beacon. Other MPs receive the stream, so they include similar elements into their beacons. The total length of a successful beacon transmission or a beacon collision is approximately 7 or 34 slots, correspondingly. We discuss the accuracy of this approximation in the end of this section. TABLE UMERICAL VALUES Slot = aslottime, us 9us acwmi slots 5 K = +*acwmin 3 Lowest basic rate, Mbps 6 Beacon payload length, bytes 07 DIFS, us 34 EIFS, us 94 t s, us 47 t c, us 307 t s, rounded to the integer number of 7 slots t c, rounded to the integer number of 34 slots ATIM Window size, slots 50 00 umber of MPs 60 Figure shows the average number of successful beacon transmissions in an ATIM Window vs. the number of MPs in the mesh network. B(,M) increases with the number of MPs until the number of beacon collisions is not too big and almost all MPs transmit their beacons successfully. After some point which depends on the ATIM Window size, B(,M) goes down as the beacon collision probability increases sharply. In the numerical analysis, we calculate the average number of beacons transmitted successfully per a Beacon Interval, B(, M) = B(, +*acwmi M),
6 B(,M) 6 B(,M) 5 5 4 4 3 3 0 0 0 0 30 40 50 M = 50 M = 70 M = 00 M = 50 M = 00 Figure The average number of successful beacon transmissions in ATIM Window vs. number of MPs Figure shows the average number of successful beacon transmissions in an ATIM Window vs. the ATIM Window size. The increase of the curves is periodic and one can see that the period is approximately equal to the beacon transmission length. Arbitrary increase of the ATIM Window may give no increase in the number of successful beacon transmissions. Figure 3 shows the probability of successful beacon transmission for an MP vs. the number of MPs. To some point the probability B(,M)/ decreases slowly because the number of beacon collisions is small. For a given number of MPs, the probability of beacon collision depends on the number of virtual slots which increases with the ATIM Window size. However, the number of virtual slots cannot be more than K=+ acwmin aslottime, so after some point the probability B(,M)/ sharply goes down as the number of beacon collisions increases quickly. However long ATIM Window gives high probability of successful beaconing, it also causes high overhead because data transmissions are prohibited in ATIM Window. To implement mesh services it may be needed to guarantee that beacon of an MP is successfully transmitted within some timeout. For example, MDA mechanism [4] requires that each MP having MDA active to advertise its MDA reservations periodically. Otherwise, the reservations become unprotected from collisions or cancelled. 0 0 50 00 50 00 = 0 = 0 = 35 = 45 = 60 Figure The average number of successful beacon transmissions in ATIM Window vs. the ATIM Window size In other words, to support MDA, it is needed to guarantee that the probability of successful beacon transmission is not less than some threshold. Results shown in Figure 3 allow adjusting the ATIM Window size depending on the number of neighboring MPs. For example, if the MDA timeout is set to five Beacon Intervals, the probability of successful beacon transmission shall be greater than 0%. Figure 3 shows that in case of 30 MPs in the neighborhood, it is needed to set the ATIM Window size to about 50 slots or.3ms. If the number of MPs is 0, then ATIM Window may be two times shorter, i.e. only 70 slots or 0.6ms. Adjusting ATIM Window with the number of neighboring MPs is an effective method to guarantee successful beaconing and maintain overhead reasonable. 0.9 0.8 0.7 0.6 B(,M)/ 0.5 0.4 0.3 0. 0. 0 0 0 0 30 40 50 M = 50 M = 70 M = 00 M = 50 M = 00 Figure 3 The probability that a beacon transmission from an MP is successful vs. number of MPs M
Figure 4 compares analytical and simulation results. The simulation model is build with General Purpose Simulation System (GPSS) [8] with continuous time, i.e. the length of beacon transmission plus DIFS or EIFS is 47 or 307 microseconds, correspondingly. Figure 4 shows that the inaccuracy caused by approximation of these values to the integer number of slots, 7 and 34 slots, correspondingly, is very low. This is not surprising and may be simply explained by the fact that the beacon length is much bigger than the maximal inaccuracy of the approximation which is equal to 0.5 slots. 5 4,5 4 3,5 3,5 B(,M),5 0 5 0 5 0 Theory, M = 50 (Ts=7, Tc=34) Simulatio M = 50 Figure 4 Comparing simulation and analytical results where values are rounded to the integer number of slots which performance is heavily affected by non-beacon transmissions. 0. References [] IEEE Std 80., 999 Editio Wireless LA Medium Access Control (MAC) and Physical Layer (PHY) specificatio August 999 [] IEEE Std 80.-007, Revision of IEEE Std 80.- 999, Wireless LA Medium Access Control (MAC) and Physical Layer (PHY) specificatio June 007 [3] IEEE P80.s/D.00, Draft Amendment to Standard, ESS Mesh etworking, ovember 006 [4] IEEE P80.s/D.08, Draft Amendment to Standard, Mesh etworking, January 008 [5] L. Huang and T.H. Lai, On the scalability of IEEE 80. ad hoc networks, In Proceedings of ACM MobiHoc, 00 [6] D. Zhou, T.H. Lai, Analysis and Implementation of Scalable Clock Synchronization Protocols in IEEE 80. Ad Hoc etworks, In Proceedings of MASS, 004 [7] Yongho Seok, Alexander Safonov, Dee Denteneer, doc.: IEEE 80.-07/73r MDA Extension for Multicast/Broadcast Transmission, ovember 007 [8] http://www.gpss.ru/ [9] IEEE LA/MA Standardization Committee, http://www.ieee80.org/ 5. Conclusions In this paper, we discuss the synchronization aspects and study the beaconing algorithm of synchronized MPs in IEEE 80.s mesh networks standard draft [3]. The probability of successful beacon transmission is calculated depending on the number of MPs in the network and the ATIM Window size. On the one hand, within ATIM Window beacons are protected from collisions with data transmissions. On the other hand, ATIM Window causes overhead and should be as short as it is possible. The results of our analysis show the method of optimal adjusting the ATIM Window size depending on the number of MPs in the network. Common ATIM Window usage is an effective method to protect beacons of synchronized MPs. However, the current situation in the IEEE LA/MA Standardization Committee [9] is that 80.s draft tends to provide no mesh-wide synchronization. Authors strongly believe that mesh-wide QoS would be difficult to achieve with unsynchronized beaconing