Effective Distributed Scheduling for Parallel Workloads

Andrea C. Dusseau, Remzi H. Arpaci, David E. Culler
ACM Sigmetrics Conference on the Measurement and Modeling of Computer Systems
May 1996, Philadelphia, PA.

This page provides simulation data for the performance of Implicit Scheduling relative to coscheduling for a variety of synthetic, bulk-synchronous parallel applications. For complete details, see the postscript version of the paper.

Pick one selection from each of the three categories.


Presentation of data:
Graph and formatted data Textual data

Communication pattern:
Barrier News Transpose

Simulation experiment:


Synthetic Workload Characterisitcs
Coscheduled Characteristics Figure 2

Verification of Previous Results
Immediate Blocking Figure 3

Two-Phase Fixed-Spin
Single Context-Switch Figure 5
Two Four Eight Times Context-Switch Figure 6 (News only)

Adaptive Two-Phase Blocking
Known Load-Imbalance Figure 8 (News only)
Local Approximation of Load-Imbalance Figure 9 (News only)
Global Approximation of Load-Imbalance Figure 10

Sensitivity to the Local Scheduler
Coordinated Timers Figure 11
Round-Robin Local Scheduler Figure 12


Back to the NOW project main page.


This page is maintained by dusseau@cs.berkeley.edu.
Last updated March 1, 1996.