A comparison of distributed data parallel multimedia computing over conventional and optical wide-area networks

The research area of Multimedia Content Analysis (MMCA) considers all aspects of the automated extraction of knowledge from multimedia data streams and archives. As individual compute clusters can not satisfy the increasing computational demands of emerging MMCA problems, distributed supercomputing on collections of compute clusters is rapidly becoming indispensable. A well-known manner of obtaining speedups in MMCA is to apply data parallel approaches, in which commonly used data structures (e.g. video frames) are being scattered among the available compute nodes. Such approaches work well for individual compute clusters, but - due to the inherently large wide-area communication overheads - these are generally not applied in distributed cluster systems. Given the increasing availability of low-latency, high-bandwidth optical wide-area networks, however, wide-area data parallel execution may now become a feasible acceleration approach. This paper discusses the wide-area data parallel execution of a realistic MMCA problem. It presents experimental results obtained on real distributed systems, and provides a feasibility analysis of the applied parallelization approach.