Modern parallel computers offer parallel disk systems, which will allow many applications to match the high compute performance of Figure 2 with scaling disk I/O (input/output) performance. The software and methodology for accessing these parallel disks is still rudimentary and more experience is needed to develop this. A major initiative---the Scalable I/O Initiative---is being led by Messina from Caltech in this area. Particularly interesting is the introduction of a commercial relational database, ORACLE, on distributed memory multicomputers with the IBM SP-2 implementation particularly attractive as IBM has already such a strong presence in the (conventional computer) commercial world.
Parallel debuggers are available, and these are clear extensions from the sequential environment for the data parallel applications that dominate science and engineering simulations. Monitoring and evaluation of the performance of the computer is particularly important for parallel machines as it can signal poor decomposition and other inhibitors to good use of the machine. Other software tools can automatically decompose and distribute problems over the nodes of a parallel machine.