Performance of the I/O subsystem plays a significant role in parallel applications that need to access large amounts of data. I/O performance in such applications is expected to be scalable and balanced with respect to the communication and CPU performance. MPIIO, a part of the MPI-2 standard has many implementations. Each of the available clientside parallel architectures differ widely in their approach to achieving high performance. This thesis hypothesizes that the effectiveness of each available client-side parallel architecture differs in delivering overall parallel application performance for a given underlying file system and that increasing the performance for different workload characteristics requires different designs. This hypothesis is validated by the development of appropriate metrics and the analysis of the results, obtained from running the experiments.