The OpenWLC is a two-years project sponsored in parts by the High performance Computing Modernization Office, U.S. Department of Defense and the Office of Advanced Scientific Computing Research, U.S. Department of Energy. ORNL is managing the project, which also involves ANL, PNNL, and LaTech. The objective of the project is to create an open-source workload characterization framework, and then deploy the framework onto the DOD/DOE high-end computing systems.

Large-scale computing systems are very difficult to manage due to their sheer size and the complexity of the applications they run. In order to fully exploit their potential power, it is critical to accurately monitor the system activities (workloads) and determine the potential performance bottlenecks that limit its effectiveness. Obviously, the challenge lies in providing a performance evaluation framework that can optimally and simultaneously meet the criterions of scalability, performance, extensibility, availability and minimum intrusiveness.

The OpenWLC system is a scalable, integrated environment for systematically collecting the monitored data and applying workload characterization techniques to raw data produced by monitoring application programs. Specifically, the OpenWLC is designed to scale up to and beyond 100,000 processors. The OpenWLC’s framework employs a component-based, multi-tier, architecture to cope with large amounts of monitored data during collection, storage, visualization and analysis stages. OpenWLC is designed such that the impact on whole system applications performance is minimal (less than 1% on an average).

The OpenWLC system will be particularly useful for program tuning and performance debugging, for testing alternative configurations and for supporting performance evaluation studies as well as characterizing application behaviors (more info). Initial work on an OpenWLC prototype is in progress. Current research involves component-based middleware, workload models and data visualization techniques.

Performance Evaluation and Workload Characterization

The performance evaluation of computer system requires understanding of a system's workload. This is because the optimal performance that a computer system can achieve depends strictly on the match between workload and system characteristics.

The OpenWLC project represents our initial efforts into understanding the issues and the methodology in system performance evaluation and workload characterization. Our longer-term objective is to develop a systematic methodology to address the evaluation of current and future applications for various computing platforms. Applications range from relatively small applications running on low-cost, low-power embedded systems, to multimedia applications (video encode/decode, speech recognition/synthesis, compression) running on dedicated mobile systems, to large-scale scientific applications, databases, and decision-support systems running on high-end computing systems, such as proprietary HPC systems, COTS clusters, Grid, and Internet.

In particular, we are interested to find ways to: