Job scheduling that minimizes network contention due to both communication and I/O

Abstract

As communication and I/O traffic increase on the interconnection network of high-performance systems

Symbiotic Space-Sharing on SDSC’s DataStar System

“…Mache and Garg have focused on finding a spatial layout for concurrent jobs in a parallel space-shared machine to minimize communication and maximize access to I/O nodes for I/O-intensive jobs [17]. We address a related but different problem in considering not only the physical layout but also the sets of jobs contending for resources.…”

Using Site-Level Modeling to Evaluate the Performance of Parallel System Schedulers

“…In addition to the scheduler and user models, a complete site-level simulation may also include a machine model. This may be important because the performance of specific applications may be affected by the machine's architecture [17], or by interference from other jobs [13]. However, such detailed simulations require much more information about applications and take much longer to run.…”

Exploring the relationship between parallel application run-time variability and network performance in clusters

“…Work related to centralized cluster scheduling include communication-aware task mapping strategies [28], contiguous allocation strategies with [18], [22] or without aggressive backfilling [24], [32]. At this time however there is no evidence of combining the gathering of communication requirements or the converse (application run-time sensitivity to communication performance) with the integration of centralized scheduling and resource management systems.…”