study of the IEEE Contention-based Request Mechanism *

Transcription

1 study f the IEEE Cntentin-based Request Mechanism * Jesus Delicad, Francisc M. Delicad and Luis rzc-barbsa Institute de Investigacin en Infrmatica de Albacete (I^A) Universidad de Castilla-La Mancha (UCLM) Albacete, Spain {j delicad,franman,lrzc}dsi.uclm.es Abstract. Bradband wireless access systems ffer a slutin fr bradband access and QS-aware multimedia services thrugh a wireless medium. The IEEE standards specify the physical and medium access cntrl layers fr bradband wireless access systems as well as the varius mechanisms t meet the quality f service (QS) requirements f a wide variety f applicatins. Amng the mechanisms defined by the standards, the bandwidth request and grant mechanisms play a central rle n guaranteeing the QS required by the subscriber statins. In this paper, we undertake the study f the bandwidth request mechanisms defined in the IEEE standards. We then prpse an apprach fr reducing the verhead required by the signalling mechanisms. ur simulatin results shw that ur apprach utperfrms the mechanisms prpsed in the IEEE standards. Keywrds: TDMA/FDD Resurce Request, WiMAX. 1 Intrductin The design and adptin f bradband wireless access (BWA) systems is ne f the mst significant netwrking research and develpment activities nwadays. BWA systems prvide fixed-wireless access t SS Subscriber Statin (residential r business custmers) t internet service prvider (ISP) facilities. Their main advantages are: rapid deplyment, scalability, lw maintenance and upgrade csts. Their installatin can be particular useful in (a) very crwded gegraphical areas such as urban areas r in (b) rural areas lacking bradband wired infrastructure. The IEEE standards have been particularly designed fr BWA systems. The IEEE standards supprt a pint-t-multipint (PMP) architecture within the GHz range f frequencies, achieving abut * This wrk has been jintly supprted by the Spanish MEC and Eurpean Cmissin FEDER funds under grants "Cnshder Ingeni-2010 CSD " and "TIN "; by JCCM under prject PAI and grant 05/112, and by UCLM under prject TC Please use the fllwing frmat when citing this chapter: Delicad, J., Delicad, F. M., rzc-barbsa, L., 2007, in IFIP Internatinal Federatin fr Infrmatin Prcessing, Vlume 245, Persnal Wireless Cmmunicatins, eds. Simak, B., Bestak, R., Kzwska, E., (Bstn: Springer), pp

2 88 PWC km f distance. Different prducts have been develped in the past few years [?]. The IEEE medium access cntrl (MAC) prtcl is a centralized cnnectin-riented mechanism. Under this prtcl, the BS is respnsible f allcating the bandwidth required by the SSs. Furthermre, the BS has t classify, priritize and schedule the SSs requests. Twards this end, the BS has t timely and efficiently manage the verall available bandwidth. The signalling prtcl is therefre a central element allwing the SSs t place their requests and the BS t issue the grants t the SSs. Regarding the IEEE signalling prtcls, the standard specifies that a SS has first t request the bandwidth accrding t the needs f each ne f the individual cnnectins assciated t the SS. Hwever, the requests can be granted using ne f tw mdes: GPC {Grants Per Cnnectin) r GPSS {Grants Per Subscriber Statin). Under the GPC mde, grants are made t individual cnnectins. The GPSS mde instead grants the bandwidth t the SS withut specifying a particular cnnectin. Under the latter mde, the respnsibility f allcating the granted bandwidth is left t the SS. In this paper, we develp and analyze a new algrithm implementing the request bandwidth primitives available t the SSs. ur prpsal aims t imprve the perfrmance f the mdes prpsed by the standards in terms f three metrics, namely the verhead, thrughput and delay. The paper is rganized as fllws. Sectin 2 prvides an verview f the IEEE standard, which includes a brief descriptin f the MAC prtcl and request and grant bandwidth mechanisms. In Sectin 3, we analyze the relevant hterature clsely related t ur prpsal. Sectin 4 describes ur prpsal and in Sectin 5, we carry ut a cmparative perfrmance study f ur prpsal with respect t the mdes included in the standards. Finally, Sectin 6 cncludes the paper. 2 IEEE The IEEE standards specify the physical (PHY) and MAC layer f the air interface f interperable pint-t-multipint and ptinal Mesh tplgy BWA systems. The range f frequencies supprted by IEEE is frm 2 t 66 GHz, which includes the licensed and license-exempt bands. Line-f-sight (LS) is smetimes necessary depending n the range f frequencies. Three types f mdulatin can be used: QPSK (Quadrature Phase-Shift Keying), 16-QAM (Quadrature Amplitude Mdulatin) and 64-QAM, but nly QPSK is mandatry. The bit rate and rbustness in the presence f errrs depend n the type f mdulatin and frequency being used. 2.1 MAC layer As already mentined, the MAC layer is a centralized and cnnectin-riented mechanism, that is, the BS is respnsible f allcating the resurces and prvi-

3 Persnal Wireless Cmmunicatins 89 sining the system with QS-aware mechanisms. T perfrm this task, the SSs are required t request t the BS their needs in a timely manner. That is t say, each SS has t request t the BS the required bandwidth n a frame by frame basis. In respnse t the resurces required by all the SSs, the BS distributes the available bandwidth taking int accunt the requirements f all the utstanding SSs requests. The cmmunicatin between the BS and the SSs is realized by means f fixedlength frames, divided int tw subframes: the dwnlink and uplink subframes, whse lengths are dynamically cntrlled by the BS. The mde f peratin can be Frequency Divisin Duplexing (FDD) r Time Divisin Duplexing (TDD). The dwnlink and uplink cmmunicatins are time multiplexed by means f a Time Divisin Multiple Access (TDMA) mechanism. The dwnlink subframe starts with a Frame Start Preamble used by the PHY fr synchrnizatin and equalizatin. This is fllwed by the frame cntrl sectin, which is cmpsed f management messages. It cntains the dwnlink and uphnk maps stating the physical slts (PSs) at which bursts begin, that is t say, they cmprise the bandwidth allcatins fr the data transmissin in the dwnhnk and uplink directins, DL-MAP and UL-MAP messages, respectively. The DL-MAP cntains the addresses f the first time slts being used t cnvey the data transmitted by the BS and its crrespndent dwnlink burst prfile (Dwnlink Interval Usage Cde, DIUC). The UL-MAP cntains the specific data (Infrmatin Element, IE) which include the transmissin pprtunities, that is t say, the time slts at which each an every active SS can transmit. After receiving the UL-MAP message, the SSs transmit their data in predefined time slts as indicated in the IE. The BS scheduling mdule determines the transmissin pprtunities (les) using the bandwidth request (BW-REQ message) sent by the SSs t BS. The size f this message affects the remaining size f the dwnlink frame, that is t say, the lnger this message, the less available bandwidth fr the transmissin f data ver the dwnlink directin and vice versa. Hwever, the UL-MAP size will increase as the number f grants in the uplink directin increases. Fllwing these maps, the DCD and UCD messages indicate the physical characteristics f the physical channels. Finally, it is intrduced a TDM prtin which carries data, rganized int bursts depending n the negtiated burst prfile between the BS and the SS, and therefre different levels f transmissin rbustness. Bursts are transmitted in rder f decreasing rbustness. In the FDD case, the TDM prtin may be fllwed by a TDMA prtin used fr transmitting data t any half-duplex SSs scheduled t transmit earlier than in the frame they are receiving data. Each burst begins with the Dwnlink TDMA Burst Preamble t regain synchrnizatin. In this case, the DL-MAP message includes a map f, bth, the TDM and TDMA bursts. The structure f the uplink subframe is divided int three classes f bursts: 1. Cntentin pprtunities reserved fr Initial Maintenance. The RNG-REQ messages are transmitted by the SSs jining the netwrk.

4 90 PWC Cntentin pprtunities defined by the Request Intervals. These are reserved t cnvey the replies t the multicast and bradcast plls, in which the transmitted BW-REQ messages are placed. These messages are issued by the SSs t indicate t the BS their needs. 3. Intervals defined by Data Grant les specifically allcated t individual SSs. The bandwidth allcated fr Initial Maintenance and Request cntentin pprtunities may be gruped tgether and is always used with the uplink burst prfiles. The remaining transmissin slts are gruped by SS. The UL-MAP message, in the dwnlink subframe, grants bandwidth t specific SSs and indicates the uplink burst prfile assigned t transmit. These data are used by each SS t transmit in its assigned allcatin. These transmissins are separated by SS Transitin Gaps in rder t prperly synchrnize t the SS. 2.2 Requests A request refers t the mechanism that SSs use t indicate t the BS that they need uplink bandwidth allcatin. This request may cme as a stand-alne bandwidth request header (BW-request) r as a piggyback request. The IEEE standard uses a randm access mechanism in the uplink subframe fr transmitting the pprtunity requests t the SSs, during the request cntentin perid defined int this subframe. The BS is respnsible f establishing this reservatin perid at the beginning f each subframe. In this way, the SSs can place their reservatins (BW-requests) fr transmitting in the next subframe (r later, depending n the happening r nt f cllisins). The standard defines the truncated binary expnential backff algrithm as the mechanism fr reslving ptential cnflicts during this interval. T limit the length f the cntentin reslutin perid is ne f the main issues t be addressed. The lnger the cntentin reslutin perid, the shrter the available capacity left fr transmitting data. Accrding t the IEEE standard, the requests have t be issued n an individual basis, that is, a SS has t issue a request fr each cnnectin assciated t it. Each request has t be identified by a cnnectin identifier dented frm nw n as CID (Cnnectin IDentifier). The use f piggyback requests is ut f the scpe f this paper, because we are interested in imprving the cntentin perid in which it is nly pssible t make BW-requests using the backff mechanism. 2.3 Grants Accrding t the IEEE standards, the grants can be issued accrding t tw schemes. The first versin f IEEE standard has defined a mechanism t grant bandwidth n a cnnectin by cnnectin basis, where each cnnectin is assciated t its crrespnding SS. This mechanism has been named Grants Per Cnnectin (GPC). The grant is assciated t the cnnectin by explicitly indicating n the grant the cnnectin identifier, CID. In the secnd apprach.

5 Persnal Wireless Cmmunicatins 91 the grants are issued n a statin basis {Grants Per Subscriber Statin, GPSS). Here, the bandwidth is granted t the SS and nt explicitly t an individual CID. Each SS is then respnsible f allcating the received pprtunities amng its different types f service flws (applicatins). It is bvius that the GPC methd requires mre bandwidth t cnvey the grants messages, since it is necessary t individually indicate the bandwidth t each cnnectin. In the GPSS methd, nly ne grant per SS is sent t the cncerned SS. There is therefre a trade-flpt be cnsidered between the amunt f signalling traffic and the prcessing required fr effectively distributing the bandwidth amng the cmpeting cnnectins. 3 Related wrk Many research effrts n the perfrmance f MAC prtcls have been carried ut and have been reprted in the literature. In [?], authrs made an analysis f different parameters invlved in the truncated binary expnential backff algrithm t reslve cllisins prduced in the system. The ptimal cntentin perid has been studied in [?], accrding t the number f users under definitin f a cst functin where the channel thrughput and delay f the system have been cnsidered the tw essential metrics. Finally, the authrs have cme t the cnclusin that the ptimal size f the cntentin perid is 2M 1, where M is the number f SSs (users). In [?], the authrs have prpsed a new QS architecture fr IEEE and have btained the size f cntentin perid accrdingly t maximize the thrughput. In this case, the result is a cntentin perid equal t the number f cmpeting SSs. In bth papers, the results highly depend n the number f SSs, s the remaining uplink subframe t send data is smaller each time the number f SSs increases. An adaptive bandwidth request mechanism implemented at the SSs fr real time traffic is prpsed in [?]. In this apprach, each SS predicts the arrival f real time packets prir t their arrival and requests bandwidth in advance. In [?], the authrs have intrduced a dynamic minislt allcatin scheme. The number f required cntentin minislts in each frame perid is based n an estimate f the maximum number f data packets that can be transmitted in a frame. The study frmally analyzes the prpsal, but it des nt include numerical results. [?] intrduces a new MAC prtcl t priritize traffic accrding t the waiting times required by the different types f traffic. The requests f the different types f traffic are assigned pririties. In [?], the authrs intrduce a new algrithm, called Multi-FS-ALHA, which divides the cntentin perid int tw parts. The first is used by the SSs issuing bandwidth request fr the first time while the secnd part is used by the SSs having previusly attempted t transmit withut success. These tw parts are dynamically fixed n a frame by frame basis. In all the studies having been reprted, the authrs make use f the mechanisms defined by the IEEE standard t request and grant bandwidth. Hwever, they d nt make an in-depth analysis f the verhead. The verhead increases as the cntentin perid increases. A simple mdificatin can be used

6 92 PWC 2007 REQUESTS CWf^" CWf^" [ C/Pf^' [ CJDf^' I... 1 CIDff," 1 cwr^ 1 1 cwr" 1 CIDf^' 1 GRANTS Zl "' I CIDf^' I CJJ)f^' I I CID^^' I [] GPSS/RGPSS: c/z)^^' c/ )^*^ " cw^^-' cw^^ Fig. 1. Cmparing GPC, GPSS & RGPSS mdes regarding requests and grants. t imprve verhead in the system intrduced by these mechanisms, especially requesting mechanism defined in the standard. This mdificatin, which is explained belw, will have a clear impact n the size f cntentin perid. 4 ur prpsal As it has been described abve, the IEEE standard specifies a bandwidth request per cnnectin made by SSs utilizing the crrespnding CID. In this way, each SS has t make a request per active cnnectin, requesting the amunt f enqueued data assciated t this cnnectin. A drawback f this resurce request scheme is that the internal cllisins increase with the number f cnnectins (an internal cllisin happens when tw r mre requests f different cnnectins within a SS try t use the same cntentin slt). The ther ne is that t keep the prbability f cllisin up when the number f cnnectins increases, the cntentin perid must increase t, reducing the size f uplink frame t be used fr data transmissin. This is derived frm [?] and [?] when the number f cnnectins per SS is bigger than ne. Regarding the grant bandwidth mechanisms, the standard describes tw methds. The GPC, which grants bandwidth per cnnectin, and the GPSS, which grants per subscriber statin. bviusly the GPSS mde imprves n the use f the frame, because the verhead in dwnlink is reduced, sending nly a grant message in the DL-MAP per SS.

7 Persnal Wireless Cmmunicatins 93 In the case f using the GPSS mde t grant resurces, the BS des nt need t knw the individual request f each cnnectin, since the grants are made by SS. In this case, the BS nly needs t knw the aggregated bandwidth request f all cnnectins f each SS. T btain this infrmatin, the BS can add up the all cnnectin requests belnging t each SS. bviusly, this requires each cnnectin t send its request, increasing the amunt f request messages and therefre the verhead in the uplink. In rder t reduce the request messages in the uphnk, we prpse that each SS shuld send a single aggregated request cntaining the ttal amunt f resurces needed by all the active cnnectins in this SS. This requesting scheme, which is called RGPSS {Request and Grants Per Subscriber Statin), tries t minimize the number f requests t be sent, reducing the verhead in the uplink phase. A direct result f the reductin in the number f request messages is a decrease in the cllisin prbability in the cntentin perid, increasing the efficiency f the resurce request mechanism based n a randm access methd. This peratin mde reduces the length f the cntentin perid since it nly depends n the number f SSs. The RGPSS scheme culd be adapted t supprt classificatins f cnnectins in rder t prvide QS. In this case, each SS will make a request per type f service flw, aggregating the resurces that all the cnnectins f a service flw need. S the BS will knw bandwidth needs f each type f service flw f all SSs, and its scheduler culd priritize sme service flws r SSs against anther ne. 5 Perfrmance Evaluatin In this sectin, we carry ut a perfrmance analysis f ur prpsal. Thrughut ur study, all simulatin are cnducted using a mdel f IEEE implemented in the PNET Mdeler vll.5 tl f?l. 5.1 Scenaris In ur simulatins, we cnsider an IEEE wireless netwrk cnsisting f several SSs and a BS describing a pint-t-multipint system. All ndes perate at 28 MHz, with a symbl rate f 22.4 MBaud. All transmissins are dne using QPSK mdulatin with a bit rate f 44.8 Mbits/s. Accrdingly t the standard a frame duratin f 1 ms is used. The mde f peratin is FDD. Ideal channel cnditins are assumed, i.e., n packet crruptin is due t the wireless channel. The system is assumed t perate in a steady-state, where the number f cnnectins des nt change ver time. Each SS runs vice, vide, backgrund and best-effrt applicatins, which are mdelled as fllws. Cnstant bit-rate vice surces at a rate f 16 Kbits/s accrding t the G.728 standard are used. The vice packet size has been set t 384 bytes including RTP/UDP/IP headers, and all vice traffic are randmly activated within

8 94 PWC 2007 the interval [0,0.024] secnds, crrespnding t a generating frequency f tw cnsecutive packets. Each SS runs 37 vice cnnectins. Fr vide apphcatins, we assume H.264 variable bit-rate vide traffic. This ne is generated frm the sequence funny encded n GIF frmat at a frame rate f 30 frames/s are used. The average vide transmissin rate is arund 1.1 Mbits/s with a packet size equal t 1,064 bytes (including RTP/UDP/IP headers). This type f traffic starts within the interval [0,0.5] secnds fllwing a unifrm randm distributin. Fr each SS, we limit t ne the number f vide cnnectin per SS. Fr the backgrund and best-effrt traffics, we cnsider a Paret distributin traffic mdel with an average bit rate fr bth types f traffic f 256 Kbits/s. The packet size has been set t a 552 bytes packet, including the TCP/IP headers. The surces are activated randmly within the interval [1,1.5] secnds. Assciated t each SS, we assume six backgrund cnnectins and ten best-effrt cnnectins. Since we are interested in studying the behaviur f the system.with the new bandwidth request mechanism, all cnnectins will request bandwidth utilizing the cntentin perid. The scheduhng algrithms used by the BS and SSs is FIF {First In, First ut). Thrughut ur study, we have simulated 10 secnds f peratin f each particular scenari, cllecting statistics after a warm-up perid f 4 secnds. Each pint in ur plts is an average ver 28 simulatin runs, and errr bars indicate 95% cnfidence interval. The study is carried ut by varying the number f SSs (2, 5, 8 and 10) in rder t increase the ttal ffered traffic. The size f the cntentin perid (#Tpp) is increased by 2 frm 4 t 10 transmissin pprtunities. We evaluate the perfrmance f the GPC, GPSS and RGPSS mechanisms allwing us t fairly cmpare them. 5.2 Results In the case f the GPC mechanism, the cllisin prbability is shwn in Figure 2.(a). As expected, the cllisin prbability decreases as the number f transmissin pprtunities increases. Mrever it increases as the netwrk lad increases. As a result, the request delay, defined as the time elapsed between the creatin f a request in a SS and its arrival t the BS, is higher when the cntentin perid size is shrter r when the traffic lad ffered increases. This is clearly depicted in Figure 2.(b). Since in GPC, the requests are sent per cnnectins and the bandwidth grants are granted in the same manner; the grant delay (Figure 2.(c)), which is defined as the time elapsed between the first request f a cnnectin and its assciated grant, behaves in a similar manner t the request delay. The results fr the metrics described abve in GPSS mde are shwn in Figure 3. Frm the results, it is clear that the cllisin prbability increases as a functin f the traffic lad ffered. In the same way than in the GPC mde, if the cntentin perid size decreases, the prbability increases (Figure 3.(a)). The request delay is shwn in Figure 3.(b), similar t the GPC mechanism.

9 Persnal Wireless Cmmunicatins 95 ^ 03 ^ H a.2 ]M n i '>^ #Tpp=4 -H '. #T0pp= #Tpp=8 --^--' #Tpp=10. A-' (a) Cllisin Prbability. ^ 03 j M (IH rt.2 'Jri ' { S (a;) Cllisin Prbability. Q : (b) Request Delay (b) Request Delay. 1 Q (c) Grant Delay. 1 Q (c) Grant Delay. Fig. 2. Results in GPC mde. Fig. 3. Results in GPSS mde. it is smaller if the number f cntentin pprtunities increases, except in the case f 8 and 10 transmissin pprtunities where this trend changes. This can be explained by the fact that fr a lnger cntentin perid, mre capacity is used by the resurce request messages resulting in a reductin n the available capacity dedicated t data transmissin. This means that the BS spends mre time t serve a request. This can be clearly seen in Figure 3.(c). This increase n the time required t send a grant results n the expiratin f the request timer. This event results in turn n the retransmissin f the request. As a result f this retransmissin, the request delay increases because upn receiving the request,

10 96 PWC 2007 CD #Tpp=4-1 ^ #Tpp= f #Tpp=8 - #Tpp= A--I --^--. ^ ^ y^' I '^ #Tpp=4 ' #Tpp=6 ' #Tpp=8 #Tpp=10/ (a) Cllisin Prbability. CD [ -^^ ^. ^ ^ ^ ^ ^^ '^^ (b) Request Delay. 1 Q I #T0pp=4 #T0pp= #Tpp=8 --^-- #Tpp=10 --^-- ) (c) Grant Delay. Fig. 4. Results in RGPSS mde. the BS cnsiders that it is the riginal request. bviusly, its creatin time is the creatin time f the riginal request. In the RGPSS mechanism, the cllisin prbability increases linearly as a functin f the lad (Figure 4.(a)), regardless f the cntentin perid values. This is because in RGPSS nly a request is sent per SS, aggregating all the cnnectin bandwidth requests f the SS. Similar behavir is depicted in Figure 4.(b), where the request delay is represented. Nt big difference culd be appreciated in the case f 6, 8 and 10 cntentin pprtunities, and nly in the case f using 4 pprtunities fr cntentin, the delay is higher than 1 ms, which culd be cnsidered as acceptable since it matches the perid f a MAC frame. The grant delay, shwn in Figure 4.(c), presents a similar trend. It increases with the injected traffic and there are nt big differences in the use f lnger cntentin perids. We nw cmpare the perfrmance f the three request and grant mechanisms. Accrding t the previus results, the best results limiting the length f the cntentin perid fr each mechanism were: 10 transmissin pprtunities fr the GPC; 8 fr GPSS and 6 fr RGPSS. Figure 5.(a) depicts the cllisin prbability. In it, the verall best result is prvided by the RGPSS mechanism, prving effective the reductin f requests sent by the SSs.

11 Persnal Wireless Cmmunicatins (a) Cllisin Prbability. JS 0.15 Q <p (b) Request Delay (c) Grant Delay (d) Thrughput. Fig. 5. Cmparisn between GPC, GPSS k RGPSS. Regarding the request delay and grant delay (Figures 5.(b) and 5.(c), respectively), the difference between the mechanisms under study is clearly appreciated. This is particularly true as the netwrk lad is increased. The RGPSS exhibits the best behavir due t the limited number f requests. As a result n the reductin in the grant delay in RGPSS, this mechanism presents a higher thrughput in uphnk directin, as seen in Figure 5.(d). This results in a lwer cllisin prbability, mre requests arrive t the BS resulting n a shrter request delay. The BS is able t grant mre cnnectins in a shrter perid f time. ther reasn fr this behavir is that cntentin perid size in the RGPSS mechanism is shrter. In cnclusin, mre capacity is available t deliver actual data. 6 Cnclusins In this paper, a new mechanism t request in IEEE netwrk, called RGPSS {Requests and Grants Per Subscriber Statins), has been prpsed. It sends nly ne request per SS, carrying the aggregated bandwidth requirements f all its assciated cnnectins. It has been shwn that the prpsal exhibits a lwer cllisin prbability in the cntentin perid and a higher thrughput in the uplink directin than

12 98 PWC 2007 the schemes defined in the standard. These results demnstrate that with the RGPSS mechanism the netwrk can supprt mre traffic. It has als been shwn that the cntentin perid size used by RGPSS is shrter than the ne required by the ther tw mechanisms. This mechanism culd be easily adapted t supprt different types f traffic (cnnectins). It culd be used as a means t prvisin a netwrk with QS supprt. In future wrks, we plan t evaluate the perfrmance f such mechanisms and the impact f ther parameters heavily impacting the length f the cntentin perid.