Services for Modeling Metasystem Performance using Queueing Network Models

The use of metacomputing systems, or metasystems, for high performance computing has gained considerable acceptance. Performance models help to obtain the ex pected app lication performance in such kind of environments. Hence, it is important that the metasystem provides services for building performance models. These services s hould support the main a ctivities related to model development: measurements to parameterize the model, and the possibility of reproducing scenarios, which is useful to the c alibration and validation tasks. Our research focuses on identifying these services and incorporating them in the SUMA metasystem. In this article, we describe the identified services and how they are going to be included in SUMA. The services will provide performance parameters, which will be used to feed queueing n etwork models (QNM). Additionally, we ex plain the implementation of the service for measuring the metasystem overhead and show some preliminary results.