Modeling "Just-in-Time" Communication in Distributed Real-Time Multimedia Applications

The research area of multimedia content analysis (MMCA) considers all aspects of the automated extraction of new knowledge from large multimedia data streams and archives. In recent years, there has been a tremendous growth (in data and computational demands) in the MMCA domain, and this growth is likely to continue in the near future. Multimedia applications operating in real-time environments must run under very strict time constraints, e.g., to analyze video frames at the same rate as a camera produces them. To adhere to such constraints, large-scale multimedia applications typically are being executed on Grid systems consisting of large collections of compute clusters. In services-based scenarios, where video content analysis is being performed by a set of remote multimedia servers, results on a particular video frame are obtained quickest if a server is unoccupied (i.e., not working on previously submitted frames). Keeping a server unoccupied, however, is a waste of available compute resources. Therefore, it is important to tune the transmission of newly generated video frames to the occupation of remote servers. However, due to variations in transmission latencies, it is difficult to accurately tune the sending of video frames such that resource utilization is optimized. In this paper we refer to this issue as the problem of "just-in-time" communication. In this paper we address this issue by introducing an adaptive control method that reacts to the continuously changing circumstances in Grid systems so as to obtain the highest service utilization possible, and to minimize service response time for individual video frames. Extensive experimental validation on a real distributed system, in combination with a trace-driven simulation, show that our control method indeed is highly effective.