22
A Case for a Mobility A Case for a Mobility Based Admission Based Admission Control Policy Control Policy Shahram Ghandeharizadeh Shahram Ghandeharizadeh 1 , Tooraj Helmi , Tooraj Helmi 1 , Shyam Kapadia , Shyam Kapadia 1 , Bhaskar , Bhaskar Krishnamachari Krishnamachari 1,2 1,2 1 Computer Science Dept Computer Science Dept 2 Electrical Engineering Electrical Engineering Dept Dept University of Southern California University of Southern California University of University of Southern California Southern California Los Angeles 90089 Los Angeles 90089 Los Angeles 90089 Los Angeles 90089 { shahram,thelmi,kapadia,bkrishna}@usc.edu shahram,thelmi,kapadia,bkrishna}@usc.edu

A Case for a Mobility Based Admission Control Policy Shahram Ghandeharizadeh 1, Tooraj Helmi 1, Shyam Kapadia 1, Bhaskar Krishnamachari 1,2 1 Computer

Embed Size (px)

Citation preview

A Case for a Mobility Based A Case for a Mobility Based Admission Control PolicyAdmission Control Policy

Shahram GhandeharizadehShahram Ghandeharizadeh11, Tooraj Helmi, Tooraj Helmi11, Shyam Kapadia, Shyam Kapadia11, Bhaskar , Bhaskar KrishnamachariKrishnamachari1,21,2

11Computer Science Dept Computer Science Dept 22Electrical Engineering DeptElectrical Engineering Dept University of Southern California University of Southern California University of Southern California University of Southern California

Los Angeles 90089 Los Angeles 90089 Los Angeles 90089 Los Angeles 90089

{{shahram,thelmi,kapadia,bkrishna}@usc.edushahram,thelmi,kapadia,bkrishna}@usc.edu

OutlineOutline What is admission control?What is admission control?

Admission control in C2P2Admission control in C2P2

Architectural FrameworkArchitectural Framework MAD (MAD (MMobility prediction based obility prediction based ADADmission mission

control policy)control policy) ResultsResults Conclusions & Future work Conclusions & Future work 

Admission ControlAdmission Control A request for a resource (usually data)A request for a resource (usually data)

To admit or not to admit….that is the questionTo admit or not to admit….that is the question Without disturbing the existing requests being servicedWithout disturbing the existing requests being serviced

Who decides?Who decides? Client, Server, Negotiation between the twoClient, Server, Negotiation between the two

After how long is the decision made?After how long is the decision made? If enough resources are availableIf enough resources are available

ADMITADMIT ElseElse

REJECTREJECT A certain level of A certain level of QoS is guaranteedQoS is guaranteed

End to End Delay End to End Delay BandwidthBandwidth Packet delivery ratioPacket delivery ratio Packet loss ratioPacket loss ratio Others….Others….

C2P2C2P2

C2P2 (Car-to-Car-Peer-to-Peer) NetworkC2P2 (Car-to-Car-Peer-to-Peer) Network Example applications video-on-demand, Audio-on-demand etc.Example applications video-on-demand, Audio-on-demand etc.

C2P2 deviceC2P2 device

C2P2 device roles: C2P2 device roles: Data producer (server), Data forwarder (router), Data consumer (client)Data producer (server), Data forwarder (router), Data consumer (client)

Admission Control in C2P2Admission Control in C2P2 ChallengesChallenges

No central coordination pointNo central coordination point Multiple simultaneous clip downloadsMultiple simultaneous clip downloads

Resource constraintsResource constraints Hiccups (jitters)Hiccups (jitters) Startup latencyStartup latency

MobilityMobility Dynamic network connectivityDynamic network connectivity Resource availability predictionResource availability prediction

FrameworkFramework A request requires all blocks of clip X be A request requires all blocks of clip X be

downloaded within Tx time units.downloaded within Tx time units. Tx = maximum tolerable startup latency Tx = maximum tolerable startup latency Hiccup free displayHiccup free display E.g. Audio clip X E.g. Audio clip X

Display time = 12 minutes Display time = 12 minutes Download time = 1 minute Download time = 1 minute

3 classes of requests3 classes of requests K = A + RK = A + R A = AA = Af f + A+ Agg

K= total requestsK= total requests R= total rejected requestsR= total rejected requests AAg g = total admitted requests serviced successfully= total admitted requests serviced successfully AAff = total admitted requests that failed = total admitted requests that failed

Example illustration of Im Example illustration of Im and Rmand Rm

SS

CC

NotationNotation: C-: C-Client,Client, S- S-ServerServer

T = 3 minutesT = 3 minutes

Tx = 1 minutesTx = 1 minutes Td = 12 minutesTd = 12 minutes Clip size ~ 30MBClip size ~ 30MB

BBdownloaddownload = 30MB/1 minute = 4Mbps < Bmax=10Mbps = 30MB/1 minute = 4Mbps < Bmax=10Mbps

=> ADMIT REQUEST=> ADMIT REQUEST

Framework (contd)Framework (contd) Request Metric (RRequest Metric (Rmm))

Amount of resources required for a given Amount of resources required for a given requestrequest

RRmm = 100 * Size(X)/(T = 100 * Size(X)/(Txx*B*Bmaxmax))

Information Metric (IInformation Metric (Imm)) Depends on the state of the networkDepends on the state of the network

0<= I0<= Imm , R , Rmm <= 100 <= 100 If (IIf (Imm – R – Rmm > > ΘΘ))

AdmitAdmit

ElseElse RejectReject

-100<= -100<= ΘΘ <= 100 <= 100

MADMAD Client-centric decentralizedClient-centric decentralized

IImm = = ΣΣi=1i=1SS T(s T(sii,C2P2,C2P2dd)) -----------------------------*100-----------------------------*100 NN^̂Tx Tx

Notation:Notation: N^ – total nodes reachable from client via probeN^ – total nodes reachable from client via probe ssii – a candidate server – a candidate server S – total list of all serversS – total list of all servers C2P2C2P2dd – C2P2 client making the request – C2P2 client making the request T – Time for which sT – Time for which si i and C2P2and C2P2dd are in radio range are in radio range

Mobility prediction based ADmission control policyMobility prediction based ADmission control policy ConservativeConservative

Typical scenarioTypical scenario

SSCC

NotationNotation: C-: C-Client,Client, S- S-ServerServer

Typical scenarioTypical scenario

Typical scenarioTypical scenario

Utility modelsUtility models UUMM = w = wRR(M)R + w(M)R + wAgAg(M)A(M)Agg + w + wAfAf(M)A(M)Aff

Experimental EnvironmentExperimental Environment

3 servers serving 100 different titles3 servers serving 100 different titles 13 Clients (1 per road stretch)13 Clients (1 per road stretch) 10 Mbps per C2P2 device10 Mbps per C2P2 device Client speed = 5 m/sClient speed = 5 m/s Reach the other end switch direction and move in the Reach the other end switch direction and move in the

opposite directionopposite direction Clients chosen randomly once every 60 secondsClients chosen randomly once every 60 seconds 100 total requests in the experiments100 total requests in the experiments

Averaged over 10 random seedsAveraged over 10 random seeds

NoteNote: : Custom simulator written in C# uses DSR as the routing protocol of Custom simulator written in C# uses DSR as the routing protocol of choice.choice.

Road stretchesRoad stretches

Utility ResultsUtility Results

Standard ModelStandard ModelU = AU = Ag g - A- Aff

Results (contd)Results (contd)

Satisfied Requests (ASatisfied Requests (Agg)) Unsatisfied Requests AUnsatisfied Requests Aff

ObservationsObservations MAD outperforms no-admission control MAD outperforms no-admission control

case in case of all the modelscase in case of all the models Choice of Choice of ΘΘ is a function of the utility model is a function of the utility model

ΘΘ = -20 = -20 is conservative is conservative No unsatisfied requestsNo unsatisfied requests Constant positive utility with all modelsConstant positive utility with all models Good for standard and premium modelsGood for standard and premium models

Related workRelated work

Admission Control in wired networksAdmission Control in wired networks Erlang B,C Erlang B,C

QoS in MANETsQoS in MANETs Part of routing protocol Part of routing protocol

INSIGNIA[4], CEDAR[6]INSIGNIA[4], CEDAR[6] Application layer QoS [5, 7] over routing Application layer QoS [5, 7] over routing

Mobility predictionMobility prediction NonStop[8]: Dynamic Data replicationNonStop[8]: Dynamic Data replication

ConclusionsConclusions

MAD performs orders of magnitude better than

the no admission control case Hence admission control is needed in a

C2P2 environment

Optimal threshold for MAD depends on the utility model

Future workFuture work

Other scenarios Mobile Servers Clients with random speed

Enhance MAD Consider available bandwidth at the servers

Streaming of clips in C2P2 Hic-cup rate

Other topologies like grid etc. Data placement & delivery scheduling issues Replication issues

ReferencesReferences [1] S. Ghandeharizadeh and T. Helmi, [1] S. Ghandeharizadeh and T. Helmi, An Evaluation of Alternative Continuous Media An Evaluation of Alternative Continuous Media

Replication Techniques in Wireless Peer-to-Peer NetworksReplication Techniques in Wireless Peer-to-Peer Networks. . Third International ACM Third International ACM Workshop on Data Engineering for Wireless and Mobile Access (MobiDE), September 2003.Workshop on Data Engineering for Wireless and Mobile Access (MobiDE), September 2003.

[2] S. Ghandeharizadeh, B. Krishnamachari, S. Song, [2] S. Ghandeharizadeh, B. Krishnamachari, S. Song, Placement of Continuous Placement of Continuous Media in Wireless Peer-to-Peer NetworksMedia in Wireless Peer-to-Peer Networks.. IEEE Transactions on Multimedia, April 2004 IEEE Transactions on Multimedia, April 2004

[3] E. Knightly and S. Shroff, [3] E. Knightly and S. Shroff, “Admission Control for Statistical QoS: Theory and “Admission Control for Statistical QoS: Theory and PracticePractice”, ”, IEEE Network, vol. 13, no. 2, pp. 20-29, 1999IEEE Network, vol. 13, no. 2, pp. 20-29, 1999

[4] S. Le, G. Ahn, X. Zhang and A. Campbell, [4] S. Le, G. Ahn, X. Zhang and A. Campbell, INSIGNIA: An IP-based Quality of INSIGNIA: An IP-based Quality of Service Framework for Mobile Ad Hoc Networks.Service Framework for Mobile Ad Hoc Networks.,,Journal of Parallel and Distributed Journal of Parallel and Distributed Computing, 60(4):374-406, 2000Computing, 60(4):374-406, 2000

[5] H. Xiao, W. Seah, A. Lo and K. Chua, [5] H. Xiao, W. Seah, A. Lo and K. Chua, A Flexible Quality of Service Model for A Flexible Quality of Service Model for Mobile Ad-Hoc NetworksMobile Ad-Hoc Networks, , IEEE VTC 2000IEEE VTC 2000

[6] P. Sinha, R. Sivakumar and V. Bhargavan, [6] P. Sinha, R. Sivakumar and V. Bhargavan, CEDAR: A Core-Extraction Distributed CEDAR: A Core-Extraction Distributed Ad Hoc Routing AlgorithmAd Hoc Routing Algorithm, INFOCOM 1999, pg 202-209, INFOCOM 1999, pg 202-209

[7] E. Pagani, G. Rossi, [7] E. Pagani, G. Rossi, A Framework for Admission Control of QoS multicast traffic A Framework for Admission Control of QoS multicast traffic in mobile Ad Hoc Networksin mobile Ad Hoc Networks, Fourth, Fourth International ACM Workshop on Wireless Multimedia International ACM Workshop on Wireless Multimedia 2001, pg 2-11.2001, pg 2-11.

QuestionsQuestions

THANK YOUTHANK YOU