COMPAS - Computational Models for Parallel Systems
The COMPAS (COmputational Models for PArallel Systems) project
formulates and studies models for various parallel and distributed
systems. Lines of research include:
- Networks: We developed and studied a distributed data
streams model, relevant to IP network monitoring devices.
- Disks: We developed models for storage systems comprised
of components containing multiple disks connected to a SCSI bus, and
improving the I/O throughput of such systems (further
details and papers).
- Multiprocessors: We developed and studied new models for
algorithm design on a variety of parallel machines; these models
include the QRQW PRAM model, the (d,x)-BSP model, and the QSM
model. We also considered runtime scheduling of program tasks to
processors for programs written in the popular data-parallel
programming model. We devised the first scheduler to guarantee load
balance among the processors while maintaining memory usage on par
with a sequential implementation. We then extended this work to
provide the first effective scheduler for programs with
synchronization variables.
Phillip
B. Gibbons Home Page