Design of Scalable Continuous Media Servers with Dynamic Replication

Multimedia applications place high demands for quality-of-service (QoS), performance, and reliability on systems. These stringent requirements make design of cost-effective and scalable systems difficult. Therefore efficient adaptive and dynamic resource management techniques in conjunction with data placement techniques can be of great help in improving performance, scalability and reliability of such systems. In this paper, we first focus on data placement. In the recent past, a great deal of work has focused on "wide" data striping as a way of dealing with load imbalance problems caused by skews in data access patterns. Another approach to dealing with load imbalance problems is replication. The appropriate compromise between the degree of striping and the degree of replication is key to the design of scalable continuous media (CM) servers. In this work we focus on evaluation of this compromise in the context of a hybrid CM server design. Changes in data access patterns lead to other questions: (1) when should the system alter the number of copies of a CM object, and (2) how to accomplish this change. We address (1) through an adaptive threshold-based approach, and we use dynamic replication policies in conjunction with a mathematical model of user behavior to address (2). We do this without any knowledge of data access patterns and with provisions for full use of VCR functionality. Through a performance study, we show that not only does the use of this mathematical model in conjunction with dynamic resource management policies improves the system's performance but that it also facilitates reduced sensitivity to changes in:(a) workload characteristics, (b) skewness of data access patterns, and (c) frequency of changes in data access patterns. We believe that not only is this a desirable property for a CM server, in general, but that furthermore, it suggests the usefulness of these techniques across a wide range of continuous media applications. (Cross-referenced as UMIACS-TR-2001-21