Analysis and Enhancement of Bandwidth Request Strategies in IEEE Networks

Transcription

1 Ths ull text paper was peer revewed at the drecton o EEE Communcatons Socety subject matter experts or publcaton n the EEE CC proceedngs nalyss and Enhancement o Bandwdth Request Strateges n EEE 8.6 etworks Chn-Tser Huang *, Chang-Lng Huang "avd Chuck", J. Morrs Chang * ept. o Computer Scence and Engneerng Unversty o South Carolna huangct@cse.sc.edu bstract EEE 8.6 based broadband wreless access network s consdered as one o the most promsng wreless access technologes. t employs a request/grant scheme n bandwdth allocaton where each subscrber staton (SS) can send bandwdth requests (BRs) to the base staton (BS) beore bandwdth can be granted to SSs. The 8.6 standard denes two types o BR strateges, namely ncremental requests and aggregate requests. n ths paper, we rst present an analytc model to analyze ther perormance and show that both types o BRs have rooms or mprovement when acng two common types o trac patterns, called the uphll trac pattern and the perodc/bursty trac pattern, n terms o overhead and data watng tme n the queue. Then, we propose two enhancement strateges, called the aggressve strategy and the conservatve strategy, that explot the property o the two trac patterns to mprove the perormance. The smulaton results show that the two enhancement strateges can eectvely reduce the overhead and data watng tme, at the prce o somewhat less bandwdth utlzaton n the case o conservatve strategy. ndex Terms EEE 8.6, WM, Bandwdth Request, Bandwdth llocaton, Trac Pattern.. TROUCTO EEE 8.6 Worldwde nterpretablty or Mcrowave ccess system (WM) s a promsng broadband wreless access technology amed to support the hgh transmsson rate and large servce range. EEE 8.6 standard [, ] denes both the Medum ccess Control (MC) and Physcal (PHY) layers o a broadband wreless network. bandwdth reservaton scheme s used n the MC layer to allow the SS to reserve any requred bandwdth rom the BS n order to support qualty o servce (QoS) guaranteeng servces. The BS coordnates bandwdth reservatons or all data transmssons and receptons. The EEE 8.6 standard denes an abstract concept o connecton and uses t to unquely denty a servce low rom an SS. The BS assgns a connecton (C) to each connecton or management purpose. The standard also speces bandwdth request/grant mechansms, whch are used to address the dynamc trac pattern and bandwdth need o a servce low over the tme. ccordng to the standard, SSs can request bandwdth or ther connectons to ensure ther QoS based on the Cs. The BS can ether grant or reject the requests based on ts avalable resources and schedulng polces. Two types o bandwdth requests (BRs), ncremental requests and aggregate requests, are dened n the EEE 8.6 standard. The ncremental request allows an SS to specy the extra amount o bandwdth needed or a specc ept. o Electrcal and Computer Engneerng owa State Unversty {chuang, morrs}@astate.edu connecton. Thereore, the amount o reserved bandwdth or ths connecton can only be ncreased by makng an ncremental request. On the other hand, the aggregate request allows an SS to specy the current state o queue or a partcular connecton. ter recevng t, BS resets ts percepton o the correspondng connecton s bandwdth needs. Consequently, the amount o reserved bandwdth or the connecton may be ncreased or decreased accordng to the state o the queue. However, although the standard speces the unctonaltes o these two types o BRs, t does not menton ther mplementatons and eects. Potentally, there are three possble mplementatons o usng BRs: (a) ncremental BRs only, (b) aggregate BRs only, and (c) both ncremental BRs (to request more BW) and aggregate BRs (to release unused BW). n ths paper, we wll ocus on cases (a) and (c), snce (b) s less ecent than (c) n sendng BR messages. Our contrbuton n ths paper s twoold. Frst, we evaluate both case (a) and case (c) n terms o bandwdth request overhead and bandwdth utlzaton. To the best o our knowledge, ths s the rst attempt to analyze those two cases o BR mplementaton. Second, we propose two enhancement strateges to mprove the perormance o the two types o BRs when acng two gven types o trac patterns, namely the uphll trac pattern and the perodc/bursty trac pattern. The rest o the paper s organzed as ollows. n Secton, we gve a bre overvew o related works. n Secton, we ntroduce our system model and assumptons. n Secton V and V, we gve analyss o the above two cases (a) and (c), respectvely. n Secton V, we propose schemes or mprovng the perormance o bandwdth request accordng to two specc types o trac patterns. n Secton V, we dscuss our smulaton setup and show the smulaton results. Fnally, we conclude and dscuss uture work n Secton V.. RELTE WORKS n the lterature, there are ew related works about bandwdth requests n WM networks. n [3] and [8], analytcal models or BR contenton perod sze are presented n order to get optmal system perormance. n [4], the authors model a contenton-based bandwdth request scheme called truncated bnary exponental backo (TBEB) and analyze ts perormance on bandwdth ecency and channel access delay. n [5], the authors propose an adaptve scheme or an SS to send the BS requests or extra bandwdth beorehand by speculatng the real tme pollng servce (rtps) trac patterns. However, none o the above works dscuss and analyze the //$6. EEE

2 Ths ull text paper was peer revewed at the drecton o EEE Communcatons Socety subject matter experts or publcaton n the EEE CC proceedngs two types o BRs (.e., the ncremental BR and the aggregate BR), ther mplementaton, and ther eects, whch are the ocus o ths paper.. SYSTEM MOEL SSUMPTOS n ths secton, we rst present the trac generaton model [6]. We assume that the real-tme trac s used wth varable bt rate (VBR) and the tme nterval between consecutve packet arrvals s consdered as exponental dstrbuton. Let λ and λ be the mean data arrval rate and the mum data arrval rate, respectvely. The steady state probablty o the trac model can be characterzed by Posson dstrbuton. We assume there are n SSs served by one BS and, or smplcty, there s only one real tme pollng servce (rtps) runnng n each SS. ccordng to the standard, the SS whch wants to set up rtps connectons wll need to negotate wth the BS the mnmum reserved trac rate (R mn ) and mum sustaned trac rate (R ) durng connecton establshments. n our model, we assume that the BS ntally assgns bandwdth B to each connecton and supports ncremental request or each connecton untl the amount o assgned bandwdth reaches R. Suppose represents the rame duraton and W s the ntal capacty per rame (n terms o byte), we can derve: W = B () ddtonally, we assume that the queue capacty o each connecton s nnte but a threshold, T, s set or ensurng the satsacton the mum delay requrement,.e. the SS tres to send BRs the amount o accumulated data n queue s greater than T. ssume s the mum tolerable delay o the connecton, then we can get the ntal threshold T and the threshold at rame as: T W = W, T = () n our model, we assume that SSs can always use ther receved bandwdth to send BRs va pggybackng n a rame wthout pollng opportuntes. s mentoned beore, ths paper attempts to analyze two cases o bandwdth request mplementaton: ) the case o usng only ncremental BRs; ) the case o usng both types o BRs. We assume that SSs use ncremental requests to request more bandwdth and use aggregate requests to release unused bandwdth. n ths paper, we do not consder the ssue o sel-correctng nature o the request/grant protocol or aggregate requests. Moreover, we assume that the arrved data are rst stored n the queue and wll be transmtted ater the prevous data n the queue has been transmtted n the ollowng rames. V. LYSS OF USG CREMETL BWTH REQUEST OLY Ths secton presents the case n whch only ncremental BRs are nvolved. s descrbed beore, the amount o data accumulated n queue s more than T, an ncremental BR wll be sent. Suppose - represents the amount o data arrved at rame - (n terms o byte), where and s the total number o rames we analyzed. The condton or sendng an ncremental BR n rame s: Q (3) + W > T where Q - s the data stored n queue beore transmttng rame -, ( - W - ) represents the amount o data n rame - that wll be added to the queue, and T - represents the threshold at rame -. From () and (3), we can derve: > W + W Q = Y (4) So, we can derve the probablty, P, o sendng an ncremental BR at rame : Ρ ( ) = λ p( (5) Y where p() s the probablty o data arrval rate beng. Suppose S s the set o rames wth uncast pollng opportuntes. The expected value o the amount o requested bandwdth, W extra, n rame s: extra λ Y = λ Y ( + M ( + M p S W (6) S where M and M P are the sze o stand-alone ncremental BR messages and the sze o pggyback subheaders, respectvely, and () represents (-Y)p(). Snce the BS can support the bandwdth need o each connecton untl reachng R, the expected granted bandwdth n rame +, W +, s mn{w +W extra (), R }. Moreover, the overhead, O, caused n rame can be derved as ollows: M Ρ O = M PΡ S S Consequently, the total overhead, O total, n rames s O total = O = The bandwdth utlzaton s an essental metrc n measurng the perormance o networks. Only when the data n queue s less than the granted bandwdth, wll the bandwdth not be ully utlzed. Consequently, we can get the ollowng condton: Q < (7) (8) + W W (9) We can rewrte (9) as: = < W + W Q Z () The expected value o unused bandwdth n rame, E, s equal to: E Z = ( Z ) p( () Thus, the bandwdth utlzaton n rame, U, s U W E = () W and the overall bandwdth utlzaton n rames, denoted by U, s U = = ( W E ) W (3) =

3 Ths ull text paper was peer revewed at the drecton o EEE Communcatons Socety subject matter experts or publcaton n the EEE CC proceedngs V. LYSS OF USG BOTH TYPES OF BWTH REQUEST ext, we evaluate the case o usng both ncremental BRs and aggregate BRs. ote that accordng to the EEE 8.6 standard, aggregate requests can only be transmtted va standalone messages. t can be done by usng ether () uncast pollng opportuntes or (B) bandwdth stealng (a term used n EEE 8.6 standard) when there s enough unused bandwdth, but no pollng opportuntes.. n the Frames wth Uncast Pollng Opportuntes We rst dscuss the case o the rames wth pollng opportuntes (.e. rame, S), whch can be used to send aggregate BRs to release unused bandwdth. The unused bandwdth exsts when the allocated bandwdth s greater than the actual requred bandwdth: Q + W < W (4) We can rewrte (4) as < W + W Q = H (5) The probablty o sendng an aggregate request n rame s H Ρ ( ) = p( (6) Thereore, the expected amount o bandwdth to be released s W = ( H ) p( d (7) release H ) B. n the Frames wth Enough Unused Bandwdth ext, we dscuss the case o the rames wthout pollng opportuntes,.e. rame S. The condton o sendng aggregate requests s very smlar to the one descrbed n the prevous case. However, snce there s no extra bandwdth receved rom the BS, SSs wll send an aggregate request va bandwdth stealng only there s enough unused bandwdth let. We can conclude the condton as the ollowng: Q + W < W M (8) where M ndcates the message sze o an aggregate request. Smlarly, we can rewrte (8) to get < W + W Q M = H (9) The probablty o sendng an aggregate request n rame s H Ρ( ) = p( () The expected amount o bandwdth to be released by the aggregate request n rame s: W release( ) = ( H ) p( () H By combnng (5), (6), (6), (7), (), and (), we can derve the overall requested change o the reserved bandwdth, W change, n rame as: Wchange = Wextra Ρ Wrelease Ρ( ) () Thereore, the reserved bandwdth n rame + should be mn{w +W change (), R }. The overall overhead or rame, denoted as O, s: O M = M P Ρ + M Ρ + M Ρ Ρ S S (3) The total overhead can be calculated usng equaton (8). Fnally, the bandwdth utlzaton n rame can be calculated as equaton (), and the overall bandwdth utlzaton n rames can be calculated as equaton (3). V. EHCEMET STRTEGES FOR GVE TYPES OF TRFFC PTTER lthough these two basc mplementatons o BRs support the general bandwdth needs o the SSs, they may not be ecent or two gven types o trac patterns, namely the uphll trac pattern and the perodc/bursty trac pattern. The uphll trac pattern s dened as a sequence o rames wth monotoncally ncreasng amount o data arrvng at the SS, untl a peak rate s reached. Common examples o the uphll trac pattern nclude TCP s slow start mechansm, and PP applcatons such as BtTorrent [9] n whch the downloader becomes the seed or more peers as the downloadng contnues. The bandwdth needs o the uphll trac pattern can be supported by usng only ncremental BRs because the bandwdth needs o the connecton keep growng. However, accordng to the 8.6 standard, an SS only requests bandwdth or the data that has already arrved at the SS rom users. That s, an SS wll request bandwdth n the next rame or the data arrvng n the current rame, and the BS wll grant the requested bandwdth ater the next rame. Obvously, there s latency between the rame durng whch the data arrved at the SS and the rame durng whch the data s actually transmtted. For the uphll trac pattern n whch the amount o data arrvng n each rame keeps ncreasng, the SS wll have to contnuously send ncremental BRs n every rame beore the peak s reached, and all the data wll suer rom the above latency. The perodc/bursty trac pattern s dened as a sequence o on rames durng whch new data arrves at the SS, ollowed by a sequence o o rames wth no new data arrvng, and the same cycle s repeated over and over agan. the number o consecutve on rames and the number o consecutve o rames are constant, we regard the pattern as perodc; examples nclude the network mantenance tasks and perodc reportng messages n sensor networks [7]. the number o consecutve on rames and the number o consecutve o rames are dynamc, we regard the pattern as bursty; examples nclude VoP and nstant messagng applcatons. The bandwdth needs o the perodc/bursty trac pattern can be supported by usng both ncremental BRs and aggregate BRs because the SS can release the unused bandwdth back to the BS durng an o perod, and request more bandwdth when the next on perod comes. However, the o perod s not very long, or example only a ew rames, then the SS wll have to send another ncremental BR to request bandwdth shortly ater an aggregate BR was sent. n ths case, the SS wll keep sendng the ncremental BR and the aggregate BR n turn, and the data wll suer rom the aorementoned latency.. ggressve Strategy or Uphll Trac Pattern To address the uphll trac pattern, we propose to use the aggressve strategy. the SS knows that the amount o data arrvng n the subsequent rames wll ncrease monotoncally, t can rst calculate the normal amount o bandwdth to

4 Ths ull text paper was peer revewed at the drecton o EEE Communcatons Socety subject matter experts or publcaton n the EEE CC proceedngs request usng ormulas () and (3), and apply the aggressve strategy to aggressvely request α percent more bandwdth n advance than what t should request. Thus, the amount o data arrvng n the next rame s wthn α percent more than the amount o data arrvng n the last rame, then the extra new data can be transmtted usng the extra bandwdth requested n advance, wthout the need or the SS to ssue another ncremental BR and wat or the BS to grant the requested bandwdth. the amount o data arrvng n the next rame s more than α percent more than the amount o data arrvng n the last rame, the SS can dynamcally adapt α to handle the excessve data. B. Conservatve Strategy or Perodc/Bursty Trac Pattern To address the perodc/bursty trac pattern, we propose to use the conservatve strategy. the SS knows that a specc connecton s assocated wth an applcaton whch generates perodc/bursty trac pattern, then when there s unused bandwdth n the connecton, the SS wll not send an aggregate BR rght away, but nstead the SS wll conservatvely hold the sendng o aggregate BR or k rames, where k s a postve nteger. new data rom the user arrves at the SS durng these k rames, then there s no need to request or bandwdth because the prevously granted bandwdth s stll avalable, and the new data can be transmtted mmedately n the next rame. no new data arrves durng the k rames, then the SS wll send an aggregate BR to release the unused bandwdth to the BS. There s a tradeo about the selecton o an approprate k. On one hand, k s dened too large, then the average bandwdth utlzaton wll be lower. n act, k s larger than the length o any o perod, then t s equvalent to the case o usng ncremental BRs only. On the other hand, k s dened smaller than the length o any o perod, then the unused bandwdth wll stll be released beore the next data arrves, makng the benet o the conservatve strategy lmted. V. SMULTO EVLUTO To very our analyss o the two basc mplementatons o BRs and evaluate our two enhancement strateges, we conduct varous smulatons usng the Qualnet 4.5 network smulator.. Perormance o Two Basc mplementatons o BR We rst smulate the two basc mplementatons o BRs. The parameters used n our smulaton envronment are shown n Table. ntally, the amount o bandwdth assgned to the connecton s. Then, when data arrves at the SS, the SS starts to send the ncremental BR to the BS. Followng to the 8.6 standard, we set the sze o one stand-alone ncremental BR message (M ) to be 9 bytes and the sze o pggyback subheader (M P ) to be 6 bytes. Fgure (a) and (b) show the bandwdth utlzaton o two cases descrbed beore and the average percentage o overhead occuped n the receved bandwdth per rame, respectvely. From Fgure (a), we can observe that the utlzaton n the case o usng only ncremental BRs s much lower than the case o usng both types o BR, and the same stuaton happens n the overhead shown as Fgure (b). We beleve that the case o usng only ncremental BRs results n low bandwdth utlzaton because the amount o receved bandwdth s never decreased even when SSs have no data to transmt. Ths also explans why the ncremental BRs are rarely sent, because the SS retans all the granted bandwdth. n the case o usng both types o BRs, an aggregate BR s sent once the receved bandwdth s more than the data n queue. Thereore, the overall utlzaton and overhead are much hgher than the case o usng ncremental BRs only. ddtonally, we can also get that both utlzaton and overhead n the case o usng both types o BRs are related wth the rame duraton. We beleve the reason s that SSs have more opportuntes to adjust the receved bandwdth n our smulaton tme. TBLE : PRMETERS USE SMULTO OF THE TWO BSC MPLEMETTOS OF BRS Parameters Values Smulaton tme 3 sec Channel requency.4ghz Trac type Varable bt rate Mean packet arrval nterval 5 ms nterval dstrbuton Exponental dstrbuton Maxmum tolerable delay msec Maxmum packet sze 4 bytes Packet sze dstrbuton Posson dstrbuton Modulaton OFM Frame duraton,.5, 4, 5, 6,, ms (a) bandwdth utlzaton (b) overhead per rame Fgure : Smulaton results o the two types o BR mplementaton. B. Perormance o the ggressve Strategy ext, we use smulatons to show the nadequacy o the two basc mplementatons o BRs when acng the uphll trac pattern and the perodc/bursty pattern, and evaluate the perormance o the two enhancement strateges. The smulaton or the aggressve strategy goes as ollows. The smulated trac pattern starts wth 5 bytes o data arrvng n the rst rame. For every subsequent rame, the amount o arrvng data ncreases by % untl reachng the mum o 496 bytes. Ths trac pattern lasts or a total o 5 rames. We test the aggressve strategy by requestng extra %, 5%, and % more bandwdth, respectvely, than what should normally be requested accordng to ormulas () and (3). For contrast purpose, we also run one smulaton whch uses the orgnal method n the 8.6 standard,.e. normal ncremental BRs wthout requestng or extra bandwdth. Fgure shows the results o our smulaton on the aggressve strategy. n ths smulaton, we consder the overhead as the total number o sent ncremental BRs. The overall latency s computed as ollows: rst, the latency o each rame o data s calculated as the derence between the ndex o the rame when the data s actually transmtted and the ndex o the rame when the data arrved at the SS. Then, the overall latency s computed as the sum o the latency o each rame o data. From Fgure (a), we can see that when the orgnal method n the 8.6 standard s used (denoted as ), the total number o sent ncremental BRs s 44. When

5 Ths ull text paper was peer revewed at the drecton o EEE Communcatons Socety subject matter experts or publcaton n the EEE CC proceedngs (a) overhead (b) latency Fgure : Smulaton results o the aggressve strategy. % extra bandwdth s requested n the ncremental BRs, the total number o sent BRs s stll the same. However, when 5% or more extra bandwdth s requested n the ncremental BRs, the total number o sent BRs starts to drop. When % extra bandwdth s requested n the ncremental BRs, the total number o sent BRs drops by about 5%. The result o overall latency, as shown n Fgure (b), s consstent wth Fgure (a), whch drops by about 5% when % extra bandwdth s requested n the ncremental BRs. Ths s because when extra bandwdth s requested n advance, t can be readly used to transmt the excessve data arrvng n the next rame, thus reducng the number o ncremental BRs and the latency o transmttng the newly arrved data. One pont worth notng s that n ths set o smulaton, all the granted bandwdth that was aggressvely requested s used, because the amount o arrvng data n each subsequent rame keeps ncreasng. Thereore, when the aggressve strategy s appled, the bandwdth utlzaton remans %. C. Perormance o the Conservatve Strategy The smulaton or the conservatve strategy goes as ollows. The smulated trac pattern s 4 consecutve on rames, wth 5 bytes o data arrvng n each on rame, ollowed by 4 consecutve o rames, n whch no data arrves. Ths trac pattern s repeated or a total o 48 rames. We test the conservatve strategy by holdng a to-be-sent aggregate BR or rame, rames, 3 rames, and 4 rames, respectvely. For contrast purpose, we also run the smulaton or the case o sendng the aggregate BRs wthout holdng. Fgure 3 shows the results o the conservatve strategy. The overhead and overall latency are computed the same way as n Fgure. From Fgure 3(a), we can see that when the orgnal method n the 8.6 standard s used (denoted as H), the total number o sent BRs (ncludng both ncremental BRs and aggregate BRs) s 3. When the aggregate BR s held or rame, the total number o sent BRs s stll 3, because the o perod s 4 rames and holdng the aggregate BR or only rame cannot catch the next on perod. When the aggregate BR s held or rames, the total number o sent BRs starts to drop. When the aggregate BR s held or 3 rames or more, we can see that the total number o sent BRs drops sharply to 4. ndeed, the 4 sent BRs are all ncremental BRs sent n the begnnng, and no aggregate BR s sent aterward because the conservatve strategy brdges the o perod gap between two on perods. The result o overall latency, as shown n Fgure 3(b), s consstent wth Fgure 3(a), whch drops sharply when the aggregate BR s held or more than rames. We also measure the bandwdth utlzaton when the conservatve strategy s used. When the aggregate BR s not held, the bandwdth utlzaton s around 9%. When the aggregate BR s held or,, 3, and 4 rames, the bandwdth (a) overhead (b) latency Fgure 3: Smulaton results o the conservatve strategy. utlzaton drops a lttle to 87%, 83%, 73%, and 73%, respectvely. Compared to the sharp drop n the overhead and overall latency, the benet o the conservatve strategy s apparent the BS has some spare bandwdth to use. V. COCLUG REMRKS n ths paper, we addressed the bandwdth request strateges or EEE 8.6 networks. We rst presented a model that analyzes the overhead, bandwdth utlzaton, and data watng tme n the queue, and showed that the two basc mplementatons o bandwdth requests n the EEE 8.6 standard have room or mprovement when acng an uphll trac pattern and a bursty perodc trac pattern, n terms o the overhead and the data watng tme. Then, we proposed two enhancement strateges that are desgned accordng to the nature o these two trac patterns to mprove the perormance. The smulaton results show that these two strateges can eectvely reduce the overhead and data watng tme at the prce o somewhat less bandwdth utlzaton n the case o conservatve strategy. Thereore, when the BS has spare capacty, the two strateges are benecal to the perormance o the network. n our uture work, we wll explore urther how to desgn adaptve bandwdth request strategy or other common trac patterns, and nvestgate the relatonshp and nteracton between the SS s bandwdth request strateges and the BS s bandwdth allocaton strateges. REFERECES [] EEE Standard or Local and Metropoltan rea etwork Part 6: r nterace or Fxed Boardband Wreless ccess Systems, EEE Std 8.6-4, 4. [] EEE Standard or Local and Metropoltan rea etwork Part 6: r nterace or Fxed Boardband Wreless ccess Systems, mendment : Physcal and Medum ccess Control Layers or Combned Fxed and Moble Operaton n Lcensed Bands, EEE Std 8.6e-5, 6. [3]. oha et al, Perormance evaluaton o reservaton medum access control n EEE 8.6 networks, Proceedngs o EEE nternatonal Conerence on Computer Systems and pplcatons, March 6. [4] J. He, K. Guld, K. Yang, H.-H. Chen, Modelng Contenton Based Bandwdth Request Scheme or EEE 8.6 etworks, EEE Communcatons Letters, Vol., o. 8, ugust 7. [5] R. Mukul, P. Sngh,. Jayaram,. as,. Sreenvasulu, K. Vnay, and. Ramamoorthy, n adaptve bandwdth request mechansm or QoS enhancement n WMax real tme communcaton, Proceedngs o 6 FP nternatonal Conerence on Wreless and Optcal Communcatons etworks, prl 6. [6] T. G. Robertazz, Computer etworks and Systems: Theory and Perormance Evaluaton, Sprnger-Verlag, 99. [7] L. Sang,. rora, and H. Zhang, On Explotng symmetrc Wreless Lnks va One-way Estmaton, Proceedngs o CM MobHoc'7, September 7. [8] J. Yan, G.S. Kuo, Cross-layer desgn o optmal contenton perod or EEE 8.6 BW systems, Proceedngs o 6 EEE nternatonal Conerence on Communcatons, pp.87-8, June 6. [9] BtTorrent,