Adaptive High-Performance Distributed Multimedia Computing

JADE-MM

Introduction

Multimedia data is rapidly gaining importance along with recent deployment of publicly accessible digital television archives, surveillance cameras in public locations, and automatic comparison of forensic video evidence. In a few years, analyzing the content of multimedia data will be a problem of phenomenal proportions, as digital video may produce high data rates, and multimedia archives steadily run into Petabytes of storage space. Consequently, for urgent and scientifically challenging problems in multimedia content analysis, Grid computing is rapidly becoming indispensable.

Emerging multimedia applications often must meet strict time constraints, even under variable workloads, data-dependent computation, and dynamic resource availability. Hence, multimedia Grid applications must be made variability-tolerant by way of controlled adaptive resource utilization. This raises the need for new stochastic runtime performance control methodologies that react to the continuously changing circumstances in large-scale Grid systems.

The JADE-MM project aims to develop stochastic control schemes that make time-constrained multimedia applications tolerant to the dynamics of large-scale Grid environments, and to integrate these control schemes into a software framework for large-scale multimedia content analysis. The project is divided in three complementary parts. Part one will research platform independent techniques for parallelization and optimization of multimedia applications. Part two will develop performance models and control schemes that are robust to the dynamics of Grid environments. Part three will research coordinated methods to reliable, adaptive, and efficient use of Grid resources. Throughout the project, results from all three parts will be integrated via software prototypes for immediate, application-driven evaluation.

Multi-disciplinary Approach

A key aspect of the project is that it requires a multi-disciplinary approach, involving new and existing solutions from the fields of multimedia content analysis (MMCA), mathematical modeling, and Grid computing. If successful, the project will significantly benefit all these research areas. For MMCA, JADE-MM will vastly enlarge the possibilities of its application areas, like surveillance camera deployment or access to petascale multimedia archives. Also, JADE-MM will allow the design of better MMCA algorithms, by enabling in-depth testing with colossal data sets. Also, a versatile MMCA execution environment is bringing deeper understanding in these algorithms and their efficiency.

For mathematical modeling, the project contains a variety of innovations. In particular, a challenging complication occurring in performance control models for Grid environments is that the speed of compute and network resources may change over time, which is fundamentally different from the classical performance control models where servers are autonomous entities, operating at a constant speed. This time-variability may have a dramatic impact on performance, requiring advancement of the theory way beyond the state-of-the-art. Moreover, the potential explosion of the model state space raises new challenges for the development of approximate control schemes to force control actions on-the-fly.

For Grid computing, the research will be right on the cutting edge of existing approaches. Grid application environments are slowly maturing towards being able to reliably submit and run applications. While fault-tolerance mechanisms for Grid applications are in their infancy, no viable approaches exist for fulfilling the elementary promise of Grids: let 'the Grid' automatically execute applications on the most suitable resources. We are aware that this is a very challenging problem. Focusing on the MMCA domain will limit this hard problem to a tractable complexity. We expect that, finally, our results will be applicable to other application domains or even to more generally applicable runtime environments.