170 likes | 280 Views
Dimensioning the Capacity of True Video-on-Demand Servers. Nelson L. S. da Fonseca , Senior Member, IEEE , and Hana Karina S. Rubinsztejn IEEE TRANSACTIONS ON MULTIMEDIA, OCTOBER 2005. Outline. Motivation When a VCR operation is performed, how to deal with this condition?
E N D
Dimensioning the Capacity of True Video-on-Demand Servers Nelson L. S. da Fonseca, Senior Member, IEEE, and Hana Karina S. Rubinsztejn IEEE TRANSACTIONS ON MULTIMEDIA, OCTOBER 2005
Outline • Motivation • When a VCR operation is performed, • how to deal with this condition? • Dimensioning the number of channels in an interactive VoD system • Accuracy of approximation model • Simulation results
Service Model • In a interactive VoD system (true VoD) • With both batching and piggybacking technique • Support VCR operations • Ex: Pause, rewind (REW), and fast forward (FF)
Motivation • Problems: • When a VCR operation is performed • become unsychronized with its multicast group • How to handle these events? • Goals: • maintain Qos for VCR operations • Minimize the number of requests which denied a VCR operation
When a VCR operation is performed • Will become unsychronized with its multicast group • Require a private channel to support, until resynchronization with another stream • Considerations: • Reservation of a pool of channels for VCR operations • Assure the VCR operations will not be denied • Resources are unnecessarily wasted • A batch stream is admitted if there are enough channels to handle VCR operations • No provision a pool of channels for VCR operations • No guarantee of QoS • No sources are unnecessarily wasted
Dimensioning the number of channels in a interactive VoD system • Characteristics: • The size of channels for VCR operators are changed • When a batch of users is admitted into, or leaves the system • How to determine the number of reserved channels? • Use an Erlang B queue • Minimize the probability of rejection of requests for VCR operations • Arrival of requests is followed by a poisson process • Use Zipfdistribution to show user preference
arrival rate The number of servers Required channels An Erlang B queue • Is a M/M/c/c queue • 1st‘M’: arrival according to a Poisson process • 2nd‘M’: exponential distribution of service time (Holding time) • 1st‘c’: the number of servers • 2nd‘c’: the limit on the clients in the queue • How to determine the number of reserved channels? • The flowchart is:
To determine the arrival rate • Two user states • Playback and VCR • The mean arrival rate of VCR requests is • : The mean arrival rate of VCR requests • : Number of users performing VCR operators • : rate of VCR requests per user • : probability of a user being in the playback state
To determine the mean holding time (H) • Includes: • The duration of the VCR operation • Resynchronization with another stream • In this paper, the unsynchronized stream merge with its original stream • The mean holding time is • : the holding time of a channel per VCR operation given that n operations are issued during a video display • : is the probability of a user requesting n VCR operationsduring the video display
The holding time h(n) (1/2) • Assuming • the duration of a VCR operation is tseconds • the request is issued at the sth frame
The holding time h(n) (2/2) • :Holding time of a channel conditioned only on the frame position • :maximum duration of an operation op which can occur at the sth frame • :probabilitydensity function for the duration of operation op , • :is the probability of any specific type of VCR operation(PAUSE, FF, or REW).
overestimation Accuracy of approximation model • To vary the mean arrival rate , different values of and were chosen = number of VCR operations issued per user varying from50 to 2000 and varying from of 1 to 10 • The higher the is, the closer the estimated valueis to the maximum, and then • Increases the chances that a request for a VCR operation • But merges with original stream will be impossible • Have a long holding time
Simulation results • Issues: • The number of users admitted into the system • The probability of reneging • The percentage of VCR operations denied
The number of users admitted into the system ◆For low loads (10 requests/min) ◆ For high loads (60 requests/ min) • For high loads • In a medium to high degree of interactivity • The different between a system with no pool and with a reserved pool is larger than when low loads are involved WC: with a contingency pool U: degree of user interactivity
The probability of reneging◆For low loads (10 requests/min) ◆ For high loads (60 requests/min) • The probability of reneging in a system with no pool is always less than with a reserved pool
The percentage of VCR operations denied◆For low loads (10 requests/min) ◆ For high loads (60 requests/ min)
In high load condition • Average of 25% of the channels being wasted • In low load condition • Average of 45% of the channels being wasted • As the number of contingencychannels increases • the number of channels admittingnew batches of users decrease • increasing the probabilityof reneging • decreasing the number of VCR operations denied.