We have developed an analytic performance model for multiple disks on a SCSI bus for a limited set of workloads. This model allows us to understand where the time is spent so that we can increase disk performance. And, we have been able to increase disk performance for workloads where the SCSI bus in the bottleneck by prefetching into the disk cache.
This is part of a larger project, COMPAS, that studies computational models for parallel systems.
We have also modeled a single disk on a bus; see more information. We have implemented and validated our model in both C and awk.
A patent has been applied for this work. Please contact me if you're interested in licensing the model or the prefetching technique; I would be happy to put you in touch with the proper Lucent people.