I/O for parallel systems has drawn increasing attention in the last few years as it has become apparent that I/O performance rather than CPU or communication performance may be the limiting factor in future computing systems. Large scale scientific computations, in addition to requiring a great deal of computational power, also deal with large quantities of data. At present, a typical Grand Challenge Application could require 1Gbyte to 4Tbytes of data per run. These figures are expected to increase by orders of magnitude as teraflop machines make their appearance. Although supercomputers have very large main memories, the memory is not large enough to hold this much amount of data. Hence, data needs to be stored on disk and the performance of the program depends on how fast the processors can access data from disks. Unfortunately, the performance of the I/O subsystems of MPPs has not kept pace with their processing and communications capabilities. A poor I/O capability can severely degrade the performance of the entire program. The need for high performance I/O is so significant that almost all the present generation parallel computers provide some kind of hardware and software support for parallel I/O.
At Northwestern University, we consider the I/O problem from a language, compiler, runtime and file systems support point of view. We are developing a sysstem called PASSION: Parallel And Scalable Software for Input-Output. PASSION provides runtime support and functions that can be used for checkpointing, saving visualization data, out-of-core computations etc. Many optimizations such as data reuse, prefetching, collective I/O are part of the PASSION system. Many software systems in I/O have adopted techniques developed in PASSION. As part of PASSION many compilation techniques for automatic data layout and reorganizations have also been developed. These techniques are also useful for managing caches. Other components of the PASSION project include a Portable Parallel File System (VIP-FS), integrating task and data parallelizm using parallel I/O, high-performance I/O techniques for OLAP and databases, applications and file servers for multimedia applications.