14.arch/RHK.keeton
.ls 2
.na
.LP
Evaluating Video Layout Strategies on a High-Bandwidth File Server

Kimberly Keeton

(Professor R. H. Katz)

AT&T Bell Labs and
(DOE) DE-FD03-92ER25135

Recent advances in workstation architecture, high-speed networking, and
storage technologies have enabled new kinds of multimedia systems.
One such system is a multimedia server that supports the playback of
stored video to client display devices. Video storage servers must
address the storage requirements of video files, as well as the strict
delay and delay jitter requirements on the retrieval of video data.
Video servers must also maximize the number of concurrent requests that
can be satisfied, without violating any of the clients' performance
requirements.

In addition to these concerns, video servers must be able to support a
widely varying client population, from portable computers with simple
black-and-white displays to workstations with sophisticated color
displays. Existing video storage systems may waste server resources
by using a single representation to satisfy all user requests. This one
representation must be converted using dithering and resizing algorithms
(in addition to decompression) to a representation that satisfies the
requested quality of service (QOS).

To address these issues, we draw upon the work of several Berkeley
research groups. First, we take advantage of the high concurrency,
bandwidth, and capacity provided by Redundant Arrays of Inexpensive
Disks (RAIDs), such as the prototype designed by Berkeley's RAID group
[2].

Second, we address the heterogeneity problem by storing multiple
resolutions of video on the server. Prof. Avideh Zakhor's group has
developed scalable compression algorithms which can be used to satisfy
client requests with representations that closely match the requested
QOS, requiring only decompression for playback [1]. The storage of
multiple resolutions of data may also be used to accommodate as many
playback requests as possible; in case of overload, the server may
switch the representations of some streams to use those that place
less demand on the server, the network, or both.

We are currently experimenting with several strategies for the layout
of scalably compressed data on a RAID-style disk array. We have
developed an event-driven simulator to evaluate the performance of
several layout schemes on the basis of disk system utilization, number
of concurrent users, and service time, for various levels of user
activity and qualities of service.


[1]	E. Chang and A. Zakhor, "Scalable Video Coding Using 3-D
Subband Velocity Coding and Multirate Quantization," 
Proc. ICASSP, 1993.

[2]	E. Lee, P. M. Chen, J. H. Hartman, A. L. Drapeau, E. L. Miller,
R. H. Katz, G. A. Gibson, and D. A. Patterson, RAID-II: A Scalable 
Storage Architecture for High-Bandwidth Network File Service, 
UC Berkeley Electronics Research Laboratory, Memorandum No. UCB/CSD 92/672, 1992.