Ph.D. Alumni: Stergios Anastasiadis

Reference:

Stergios Anastasiadis
Supporting Variable Bit-Rate Streams in a Scalable Continuous Media Server
Ph.D. Thesis, Department of Computer Science, University of Toronto, Toronto, Canada, 2001.
Primary supervisor: Prof. Kenneth C. Sevcik

Supervisor(s):

Kenneth C. Sevcik (primary)
Michael Stumm

Download Thesis:

PDF

Abstract:

Variable bit-rate encoding of motion video has been shown to generate streams of considerably smaller size than constant bit-rate encoding of equivalent quality. Therefore, efficient support for bit-rate variability in continuous media servers has the potential to significantly reduce the requirements for disk storage space, disk bandwidth, server buffer space, and network bandwidth. Nonetheless, such flexibility has been previously discouraged due to system design complications and excessive expectations from technological progress.

In this thesis, we describe the design of a distributed media server architecture and the implementation details of a prototype. Variable bit-rate streams are striped efficiently across multiple disks with deterministic quality of service guarantees. In contrast to results of earlier studies, the number of concurrent playbacks supported is shown to increase almost linearly with the number of disks. Several factors contribute to this conclusion, including the performance evaluation method and the resource management policies that we use. New approaches are introduced for scheduling the playback requests and disk transfers, organizing the memory buffers, allocating the storage space, and structuring the stream metadata. We justify several of our decisions with comparative performance measurements using both synthetic benchmarks and actual experiments with variable bit-rate streams.

High variability in stream resource requirements can lead to reduced utilization of system resources. Previous smoothing techniques tried to address this problem by prefetching stream data into buffer space of the client device. Thus, the maximum transfer bandwidth can be reduced depending on the memory configuration of the individual client. Instead, we introduce a new smoothing algorithm that can decrease the maximum required disk bandwidth by prefetching stream data into server buffers. We show that the algorithm has optimal smoothing effect under the specified constraints, and can be successfully applied to streams striped across either homogeneous or heterogeneous disks. Experiments with our prototype server demonstrate considerable improvement in the disk bandwidth utilization, and the number of streams supported at different system scales.

Keywords:

Media servers, variable bit rate streams, buffering, disk striping, storage systems

BibTeX:

@phdthesis(Anastasiadis-PhD01,
    author = {Stergios Anastasiadis},
    title = {Supporting Variable Bit-Rate Streams in a Scalable Continuous Media Server},
    school = {Department of Computer Science, University of Toronto},
    address = {Toronto, Canada},
    supervisors = {Kenneth C. Sevcik (primary), Michael Stumm},
    year = {2001},
    keywords = {Media servers, variable bit rate streams, buffering, disk striping, storage systems}
)