Buffered Co-Scheduling
The scheduling of parallel jobs is a challenging problem because the
performance and applicability of parallel scheduling algorithms is
highly dependent upon factors at different levels: the workload, the
parallel programming language, the operating system, and the machine
architecture. The importance of job scheduling strategies stems from
the impact that they can have on the resource utilization and the
response time of the system.
We recently proposed a new approach to job multitasking, called
Buffered Co-Scheduling (BCS). BCS exploits the positive aspects of
both explicit and implicit job coscheduling using three innovative
techniques: communication buffering, strobing, and non-blocking,
one-sided communication. The benefits of buffered coscheduling
include higher throughput, dramatic simplification of run-time
support, reduced communication overhead, efficient global
implementation of flow-control strategies and fault-tolerant
protocols, and accurate performance modeling.
