Using the SMiLE Monitoring Infrastructure to Detect and Lower the Inefficiency of Parallel Applications

The efficiency of a parallel execution is always underlined. Many parallel programs, however, don’t run efficiently upon Networks of Workstations for a variety of reasons such as frequent accesses of remote memories. In this article we present a hardware-based monitoring system which uses a hardware monitor to keep watch of all of the underlying transactions on the network and allows tools to utilize the observed information about the memory behavior of parallel programs to manipulate their execution in run-time. This makes the parallel applications execute more efficiently upon clusters.