James D. Salehi

Supporting stored video: reducing rate variability and end-to-end resource requirements through optimal smoothing

VBR compressed video is known to exhibit significant, multiple-time-scale bit rate variability. In this paper, we consider the transmission of stored video from a server to a client across a high speed network, and explore how the client buffer space can be used most effectively toward reducing the variability of the transmitted bit rate.We present two basic results. First, we present an optimal smoothing algorithm for achieving the greatest possible reduction in rate variability when transmitting stored video to a client with given buffer size. We provide a formal proof of optimality, and demonstrate the performance of the algorithm on a set of long MPEG-1 encoded video traces. Second, we evaluate the impact of optimal smoothing on the network resources needed for video transport, under two network service models: Deterministic Guaranteed service [1, 9] and Renegotiated CBR (RCBR) service [8, 7]. Under both models, we find the impact of optimal smoothing to be dramatic.

Providing VCR capabilities in large-scale video servers

Providing smooth playback capabilities for video servers, which must support potentially thousands of on-demand users, has been an area of active research. From a users perspective, VCR functions of fast-forward and rewind (FF/Rew), are desirable features in video-on-demand. But FF/Rew at n times the regular playback rate requires n times the regular playback bandwidth from architectural components of the video server. Thus, guaranteeing sufficient bandwidth to enable users to perform FF/Rew reduces the number of supportable users by a factor of n. In this paper we propose an alternative, effective FF/Rew service, which provides FF/Rew capabilities with an associated statistical quality-of-service (QoS) guarantee. This service provides immediate access to full-resolution FF/Rew bandwidth with high probability. When bandwidth is not available, service is either delayed or provided immediately but with a loss in resolution. In addition, we specify several QoS metrics to characterize the delay or loss experienced by a FF/Rew request. We show that using effective FF/Rew with statistical guarantees on these QoS metrics results in a significant increase in the number of supportable users, when compared to systems in which FF/Rew bandwidth is statistically reserved for each user. Moreover, a playback-only video server can be extended to provide FF/Rew service by reserving only a small portion of its total bandwidth, which is dynamically shared among FF/Rew requests.

The effectiveness of affinity-based scheduling in multiprocessor network protocol processing (extended version)

Techniques for avoiding the high memory overheads found on many modern shared-memory multiprocessors are of increasing importance in the development of high-performance multiprocessor protocol implementations. One such technique is processor-cache affinity scheduling, which can significantly lower packet latency and substantially increase protocol processing throughput. We evaluate several aspects of the effectiveness of affinity-based scheduling in multiprocessor network protocol processing, under packet-level and connection-level parallelization approaches. Specifically, we evaluate the performance of the scheduling technique (1) when a large number of streams are concurrently supported, (2) when processing includes copying of uncached packet data, (3) as applied to send-side protocol processing, and (4) in the presence of stream burstiness and source locality, two well-known properties of network traffic. We find that affinity-based scheduling performs well under these conditions, emphasizing its robustness and general effectiveness in multiprocessor network processing. In addition, we explore a technique which improves the caching behavior and available packet-level concurrency under connection-level parallelism, and find performance improves dramatically.

Playback restart in interactive streaming video applications

Low latency is crucial in networked multimedia applications such as on-demand streaming audio and video. In this paper we consider the problem of restarting or resuming playback following an interactive operation such as fast forward, rewind, or indexed jump in stored on-demand video. We present two approaches to restart playback after an interactive operation, and develop algorithms to compute the latencies incurred in restarting playback. Using long MPEG-1 traces, with an optimal smoothing technique to transmit data for regular playback, we demonstrate that the latencies incurred under our approaches are very small. We next examine the latencies incurred when restarting playback in a video server under two policies for sharing its bandwidth: (a) one in which only regular playback bandwidth is available to restart playback and (b) one in which additional free server bandwidth (if any) is accessible. We find playback restart latencies to be similarly low under both policies, suggesting that the simpler approach (a) is sufficient.

The performance impact of scheduling for cache affinity in parallel network processing

We explore processor-cache affinity scheduling of parallel network protocol processing, in a setting in which protocol processing executes on a shared-memory multiprocessor concurrently with a general workload of non-protocol activity. We find that affinity-based scheduling can significantly reduce the communication delay associated with protocol processing, enabling the host to support a greater number of concurrent streams and to provide higher maximum throughput to individual streams. In addition, we compare the performance of two parallelization alternatives, locking and independent protocol stacks (IPS), with very different caching behaviors. We find that IPS (which maximizes cache affinity) delivers much lower message latency and significantly higher message throughput capacity, yet exhibits less robust response to infra-stream burstiness and limited intra-stream scalability.

Scheduling for cache affinity in parallelized communication protocols

We explore processor-cache affinity scheduling of parallel network protocol processing in a setting in which protocol processing executes on a shared-memory multiprocessor concurrently with a general workload of non-protocol activity. We find that affinity scheduling can significantly reduce the communication delay associated with protocol processing, enabling the host to support a greater number of concurrent streams and to provide a higher maximum throughput to individual streams. In addition, we compare implementations of two parallelization approaches (Locking and Independent Protocol Stacks) with very different caching behaviors.