Efficient Support for Interactive Browsing Operations in Clustered CBR Video Servers
IEEE Transactions on Multimedia, Vol. 4, No.1, March 2002Min-You Wu and Wei Shu
Outline
Introduction Video server architectures The prefetching approach The grouping approach Simulation results
Implementing fast-forward
Increasing network bandwidth Displaying frames at a higher rate
Without increasing network bandwidth Reserve a separate encoded video file
for each ratio Block-skipping Frame-skipping
Server architecture Storage node
Storing video data Dealing its own disk scheduling algorith
m Delivery node
Receiving requests from clients Buffering and resequencing video blocks
from storage nodes Sending video blocks to clients
Distribution of video blocks
A video file is partitioned into many video blocks (CBR)
Video blocks are evenly distributed over N storage nodes in a round-robin fashion. Block i is stored in the disk (i mod N)
Scheduling of video blocks for normal play
Time cycle: time to play a block
Time slot: storage node can service multiple requests in a time cycle
The prefetching approach Block-skipping
When performing an fast-forward operation of ratio f, a block is retrieved after skipping f-1 blocks.
To avoid a hot spot, the number of storage nodes (N) and fast-forward ratio (f) must be relatively prime.
Otherwise, a video file can be distributed to a subset of storage nodes.
For example, if N=4 and f=2, then only storage node 0 and 2 will be accessed.
Delay The delivery node retrieves blocks in the seq
uence of 0, 6, 12, 3, 9, 15, …and it delivers blocks in the sequence of 0, 3, 6, 9, 12, 15, …
The block 0 needs to be delayed to time t2, so that at time t3 block 3 can be delivered.
Consider an fast-forward operation of ration f starts at tk. The maximum delay
kNkfNk
modmax0
Frame-skipping Skipping frames within a video block.
Base substream – contains the frames for fast-forward
Enhancement substream – contains the frames only for normal play.
The grouping approach The pace is defined as the speed of play. The grouping approach divides requests
into different groups based on their paces.
Operations
Change the membership of a request Adding Removing
Change the number of time slots allocated to a group Expanding Shrinking
Comparison Prefetching approach
Fully utilize the system bandwidth Initial delay Require some buffer space
Grouping approach Does not require extra buffer space The shrinking operation may lead to reall
ocation and delay of some requests.