Designing a multiuser HDTV storage server

Abstract

Future advances in networking coupled with the rapid advances in storage technologies will make it feasible to build a HDTV-on-demand server (that provides services similar to those of a neighborhood videotape rental store) on a metropolitan-area network. In this paper, we present a quantitative study of designing a multiuser HDTV server, and present efficient techniques for (1) storing multiple HDTV videos on disk, and (2) servicing multiple subscriber requests simultaneously, both under the constraint of guaranteeing HDTV playback rates. We develop a model that relates disk and device characteristics to the HDTV playback rate, and derive a storage pattern for HDTV video streams that guarantees their real-time retrieval. Given multiple HDTV streams, we develop mechanisms for merging their individual storage patterns together. We propose an off-line merging algorithm that can be applied a priori, and an on-line algorithm suitable for merging a new HDTV stream into a set of already stored HDTV streams, both of which yield a large improvement in space utilization over storing each of the streams independently. We study various policies, such as, round robin and quality proportional for servicing multiple subscribers simultaneously. The quality proportional algorithm retrieves video frames at a rate proportional on an average to the HDTV playback rates of subscribers, but uses a staggered toggling technique in which successive numbers of retrieved frames are fine tuned individually to achieve the servicing of an optimal number of subscribers simultaneously. The algorithm is powerful enough to accommodate bounded availability of HDTV display buffers, and permits dynamic additions and deletions of subscriber requests in a transparent manner (i.e., without causing discontinuity in the retrieval of any of the existing subscribers). In summary, our studies provide a quantitative demonstration of the technological feasibility and economic viability of HDTV-on-demand servers on metropolitan area networks.

Development of Columbia's video on demand testbed

“…Innovative methods for indexing/searching images by image contents were proposed in [5,8,16]. Dedicated storage architectures for real-time multi-access have been studied in [10,11,12,13,17]. Systematic approaches to the design of video servers (VS) are being undertaken in [3,18,20] as well.…”

Semi-continuous transmission for cluster-based video servers

“…Within this context, techniques have been proposed to address placement of media on disk to ensure real-time retrieval [2,29], admission control procedures to maximize server throughput [27] and buffer management policies to minimize memory requirements [13,17]. Replication and striping strategies for optimizing storage across disk arrays are described in [16,25].…”

Providing resource allocation and performance isolation in a shared streaming-media hosting service

“…the continuous playback requirements should never be violated for the entire service duration. The corresponding algorithms are characterized by the worst case assumptions regarding the service time from disk [6,12,14,16,17,4]. In [15,8,1,7], statistical ac-algorithms are designed which provide probabilistic QoS guarantees instead of deterministic ones, resulting in higher resource utilization due to statistical multiplexing gain.…”