Performance tuning for high performance computing systems

A Distributed System is composed by integration between loosely coupled software components and the underlying hardware resources that can be distributed over the standard internet framework. High Performance Computing used to involve utilization of supercomputers which could churn a lot of computing power to process massively complex computational tasks, but is now evolving across distributed systems, thereby having the ability to utilize geographically distributed computing resources. We often do not realize that today, we unknowingly are dependent on one or the other distributed high performance computing systems. With the world progressing towards paradigms such as cloud computing, installing software and buying high end computing systems will be a thing of the past, with everything being accessible through the internet. So much dependent on such systems will ensure their quick growth and the amount of resources under these systems will increase drastically in count as well. We are though not ready for such a development, mainly because the domain of resource utilization still lacks some control for the user and the system developer as well. This thesis therefore explores the specific domain of resource utilization across a computing grid and highlights a key aspect of resource allocation. We also describe a performance tuning application that validates our proposed hypothesis by highlighting the amount of resource savings that can be achieved if such an optimization is incorporated.