Development of a marked structure for traces of parallel and distributed systems

• Autor(a) principal: Da Silva, Diogo T. [UNESP]
• Data de Publicação: 2014
• Outros Autores: Manacero, Aleardo [UNESP], Lobato, Renata S. [UNESP], Menezes, Denison [UNESP], Spolon, Roberta [UNESP]
• Tipo de documento: Artigo de conferência
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/171583

The evaluation of high-performance systems, including grids, depends strongly of the workload applied during benchmarks or simulations. This is more evident with simulations, where the workloads may be created either by random loads or trace files. Although several models for generating random workloads have been proposed, trace files are the only form to assure reproducible simulations. Unfortunately there are very few trace files available in workload databases, and most of them have not been well maintained. Other problems include missing data fields and a structure for the data that is not easy to read and collect. Here we present a framework that allows the creation of trace files in which data is marked through XML tags, making easy their reading, and also provides front-end converters for some of the trace patterns found in the literature. It also can be used to collect traces from grid simulations performed in iSPD, a grid simulator based on iconic modeling, allowing for the reuse of the simulated workload and the filling of missing data. We present results with an implementation of this framework for iSPD, where we achieved smaller trace files and computing costs, even with the addition of markups. These results indicate that this approach could create a stronger pattern for workload trace files.