调度交互式视频点播服务器容量需求_英文_

第20卷第19期 系 统 仿 真 学 报© Vol 20No 192008年10月 Journalof System Simulation Oct

Server Capacity Requirements for Interactive VODitZhHi ISYintg-jEiani ,WiAN&GTBhin-lqiaRn&gD,LCI Yta-Znhan h

(National Digital Sw c ng ys em ng neer ng ec no ogy en er, engz ou 450002,China)

交互式视频点播服务器容量需求

智英建,汪斌强,李雅楠

(国家数字交换系统工程技术研究中心河南郑州450002)

摘 要真正的视频一点播系统(TVOD)能够使用户在任意时刻观看任意节目且支持用户任意的VCR交互但服务器

资源消耗巨大。提出 种新的支持用户操作的流媒体调度方案规则组播固定调度RMFS。 RMFS周期调度常规组

播保证了流合并中目标流的存在。分析了RMF S方案的最佳组播间隔和服务器容量需求。仿真结果证明RMF S可

扩展性好 即使在很高的客户交互强度下 与TVOD相比也能够减少服务器带宽消耗917%。 关键词视频点

播交互调度常规组播

中图分类号 TP393 文献标识码 A 文章编号 1004-731X(2008) 19-5370-04

Introduction RMFS adopts the class-basedadmission controlpolicyto admit

expensive and non-scalable Though many efficient multicast- 1 The Proposed Scheme

which may de,grade user experiences One alternative approach, 1.1 Request admission

[5] is to use dedicated contingency channels for handling VCR The request admission determines whether a requestoperations,however, this reservation causes the waste in server should be admitted by the current state of the VOD serverresources Recently,The static full stream scheduling(SFSS) [6] RMFS divides the request at the server end into three classes:is proposed to support interactive oprations through stream the Multicast Request (MR) generated by the server every Wmerging SFSS uses the simulation method to tune the mulitcast seconds, the Interactive Request (IR)generated by the admittedinterval,which is time-consuming Another potential drawback client,and the Admission Request (AR)generated by the newlyof SFSS is that it can not assure the quality of service(QoS)for arrived clientsinteractive requests when the client interactive intensity is high Let L denote the video length The server reserves

In this paper, the Regular Multicast Fixed Scheduling L/Wchannels to serve MR So the regular multicastscheme (RMFS) is proposed to support the user interactivity (sRtrSe)amcan

FRoeucneidvaetdio:n i2te0m08: -N06at-i1o9nal Grand FRuenvdiasemde:nta2l0R08es-0a8r-c1h5973 Pro ram of whole video from the beginning to the end

Binitne-rqesitanisg,fobcoursnedinon19b6r3o,adpbroafnedssionrf,oPrmhatDionsunpetewrvoirsko;r,LhiiYs cau-nrraen,t brorsneairnch class In particular, the threshold for IR and AR is denoted as l11978,M Sc,her research interest focuses on streaming media

n 1 and n2 be the number of IR and AR requests currently in average frequency at which a client generates valid interactiveservice Suppose that the server capacity used to serve IR and playback operations We use the VCR model depicted in Fig 2AR is N, one IR or AR request is admitted if and only if the Beginning at the PLAY state, the client will randomly transit tofollowing conditions are satisfied: other VCR states or remain at PLAY state according to the

1

When one IR or AR is admitted, the server should initiate a random jump distance in the forward/backward direction,one unicast stream to serve the request immediately At the which is exponentially distributed with meanµf/µb secondssame time one target regular multicast stream should be found, The valid VCR operations include PAUSE,JF and JBif any, to consolidate the unicast stream RMFS adopts the Proposition: The CII is totally determined by the givenlogical-start-time-based method to determine the target stream interactive parameters and the video length

The actual start time defines the time the movie started Proof:Let X1 and X2 denote the mean number of interactionsplaying Suppose the actual start time of one stream is tr,and the for the session without and with ABORT, respectively Let xn,xf,requested video time offset is pr relative to the beginning of the xb and xp be the number of PLAY, JF, JB and PAU during oncevideo The logical st’art time tv for this stream is defined as video session We use the similar method in [8] to find X1tv=tr−pr In Fig 1 PS2 s actual start time is t4,but the log’ical start According to the interactive model,we havetitmet itis t4−pA3,lli eth, t3 Tl hei vildteio server retictortdsthall thte TRSTs logical xn nxb( n b)xf( nf)xpnL

In Fig 1, the client c1 arrives at time t1, and caches video Assume that the arrival of the clients follows a Poisson processdata from RS 1 The missed part of the video is delivered by the with an arrival rateλWe consider the worst case in wh’ich eachdedicated unicast stream PS 1, called patching stream, to enable interactive command leads to a jump out of the user s bufferit to begin playback Assume the client issues one jump backward data and the user is completely merged with the RM before herequest at time t4 and the requested playback time offset relative issues another VCR operation Then arrival rates for IR and ARto the beginning of the video is p3 If the admission condition is are λ1=(1−PB2)λCII and λ2=λ, respectively Under thissatisfied, t’he server initiates one patching stream PS2 to continue cSiurmcumisntanuce, the umnuicltaisctastreaanmdsuonnicavsetracogme aroenoenftlsentghthe Wto/t2altheclient svideodisplay At the sametimetheserverfindsone g p p ,target RS(RS 1 in this example)based on the logical start time of serveRr=resLo/uWrc+e c1o−nPsumpλtioWn ra/2te+R1−isP

PS2 to merge this client back to RS1 The client concurrently buffers h r d ( hB2) id2 bri ( DBi1)ffλ1 Wir/2idata from RS1 and PS2 The length of PS2 is (p4−p3) seconds w ere ri enoithes t e v eo t rdate i erhent at nl g the above2 T

2.1 Client interactive intensity T

The client interactive intensity (CII) is defined as the server channel number requirement achieving the target QoS level

第20卷第19期 系 统 仿 真 学 报 Vol 20No 192008年10月 Journal of System Simulation Oct ,2008

PB1<P1 N l2=l2−1 3.3 Scalability of RMFS

EYxit TVOFDig 7 sihowt-sbthatdthebrethquired servericthapalciitytof RiMFlS atnd

3 Simulation and Numerical Results of 01, 05 and 10 reque,sts/s when C, II is equal to 20 The

and HI=(03, 03, 01, 001,µ) Set C=100,W=1200s, l1=94, Fig 8 CapaCcCI IiItIy vs CII Fig 9 ClliCienatpAarrciviviatalylRRavatsete((raererqqruiuevesstastsl/s/sr))atel2=94, λ=001requests/s, µ=300s-700s Fig 4 shows that the scalculation results fit simulation ones very well In reality,CII Fig 8 shows that the capacity requirement of RMFSis usually less than 10[9],and the calculation error is below 1% increases with CII,while the channel requirement of TVODin this case,which verifies the correctness of the analytical does not change But we can see that RMFS can reduce servermethod for CII. resource by as much as 917%compared to TVOD, even at

Fig 4 CII vs distance Fig 5 Probability vs multicast interval Wi e cl atni see thltat Acalctuhlatedb valuetsi arie tchont stihstent fwith thes mu a on resu s no er o serva on s a e per ormance

· 5372·

RMFS can reduce capacity requirements by as much as 40%at 4 ShovemWerY B L D: i H, t l :D i dhigh interactive level [ ] Alganorithm, Seupport,ing aVCR, FeunactionsesbgynEaxntendIemdplSetmreeanmtatMioenr:giAnng

4 Conclusions [J] Computer Science(S1002-137X),2005,32(3): 88-94

indteriactiive platybalck lciontrolds RMFlSid atdopths the i clatss-bt ased [6] Wong Y W, ,Lee J,Y B, Li V O K, et al Supporting Interactivea m ss on con ro po cy an conso a es e un cas s ream Video-on-Demand with Adaptive Multicast Streaming[J] IEEE Transthrough stream merging We derive the optimal multicast on circuits and systems for video technology(S 1558-2205), 2007,interval and server capacity requirement for RMFS Numerical 17(2): 129-142and simulation results show that RMFS scales well compared to [7] MDiutnrdiburt Pd,VSiodod-An-KD,mSiamndonS Rt CmlassJ-BaIEseEdEATcrcaenss Ciontritol afnodrTVOD and Patching even at high interactive intensity RMFS s stemufoer videoeotecohnoleo (S15y5s8e-22s05[)]2005 15(7):s84c4-c8u53scan achieve zero-delay service, and can be generalized to many [8] Mya H D, Shin K G Pegrfyormance Analy,sis of, the Interactivity f

References: Communication and Networks (IC3N),Arizona,USA,October 2001,

[1] Cai Y,Hua K A Sharing Multicast Videos Using Patching Streams [J] ULiSA:WIEECEh,2001 W:535-538M d

Multimedia tools and applications (S1380-7501), 2003, 21(2): [9] Stu i, - e-nDg dQ J oJ elingl Intferacltivte Ui ser&Beihafvior tin125-146 reamngon eman [ ] ourna o eecroncs norma on

[2] Ganjam A, Zhang H Internet Multicast Video Delivery [J] technology(S1009-5896),2007,29(9):2252-2256

Proceedings of the IEEE(S0018-9219),2005,93(1): 159-170

[3] Rocha M,Maia M,Cunha I, et al Scalable Media Streaming to

(上接第5364页) [9] SHI Feng,WU Dong-po,LI Qun,WANG Wei-ping Behavior Modeling

[5] A J Stewart, J F Cremer Beyond Keyfr’aming An Algorithmic Aproach Approach for Causal Tracing of Complex Simulation [J] Journal ofto Animation[C]//Graphics Interface 92(0-9695338-1-0)New York, System Simulation(S 1004-731X),2007, 19(2):405-409USA:ACM, 1992:273-281 [10] Song Liming A System of Generating 2D Animation from Video[D]

[6] M van de Panne M,Lamouret A Guided Optimization for Balan'ced Zhejiang University 2006 (in Chinese)

Locomotion [C]// Computer Animation and Simulation 95, [11] Y T Zhuang,X M Liu,Y H Pan Video motion capture using featureEurographics Animation Workshop Poitiers,France: Springer Verlag, tracking and skeleton reconstruction[C]// International conference on

1995: 165-177 image processing (0-7803-5467-2) Washington, DC, USA: IEEE

[7] Badler Norman I,Chi Diane M,Chopra Sonu Virtual human animation Computer Society, 1999:232-236based on movement observation and cognitive behavior models [C]// [12] Y T Zhuang,Q Zhu,Y H Pan,X M Liu Hierarachical model basedComputer Animation, Conference Proceedings (0-7695-0167-2) human motion tracking[C]// IEEE International Conference on ImageWashington,DC,USA: IEEE Computer Society, 1999: 128-137 Processing(S1522-4880),Vancouver BC,Canada,2000,3:706-709

[8] Ortiz A,Oyarzun D,Aizpurua I,Posada J Three-dimensional whole [13] Chen Rui Research on 3D human motion tracking based onbody of virtual character animation for its behavior in a virtual probabilistic model Institute of computing technology [D] Chineseenvironment using H-Anim and inverse kinematics [C]//Proceedings academy of sciences 2005 (in Chinese)of Computer Graphics International Conference(1530-1052),Washington, [14] Gao Yan Content-based human motion retrieval and motion synthe si sDC,USA: IEEE Computer Society 200419(6):307-310 [D] Shanghai jiaotong University 2006 (in Chinese)

(上接第5369页) Transactions B(S 1073-5615),2003,34B(2): 183-191

[10] Hatano M, Fukuda M,Takeuchi M An Experimental Study of the [13] Rajneesh S, Sarkar S,Gupta G S Prediction of Raceway Size in BlastFormation of Raceway Using a Cold Model [J] JISI (S0021-1575), Furnace from Two Dimensional Experimental Correlations [J] ISIJ1976,62(1):25-32 International(S0915-1559),2004,44(8): 1298-1307

[11] Flint P J,Burgess J M A Fundamental Study of Raceway Size in Two [14] Chen W H, Du S W,Yang T H Volatile Release and ParticleDimensions [J] Metallurgical and Materials Transactions B Formation Characteristics of Injected Pulverized Coal in Blast(S1073-5615), 1992,23B(3):267-283 Furnaces [J] Energy Conversion and Management (S0196-8904),

[12] Sarkar S,Gupta G S,Litster J D,Rudolph V,White E T A Cold 2007,48(7):2025-2033

Model Study of Raceway Hysteresis [J]Metallurgical and Materials