Ray Scheduler Prototype

Overview

Aims of the scheduler prototype include the following:

• Same APIs as a production scheduler
• Easy to add new algorithms
• Able to replay production logs
• Support unit tests for core scheduler and for database (Redis) integration

Usage Example

(NUM_NODES=3
NUM_WORKERS_PER_NODE=1
for trace in $(ls traces/*.json); do
    python replaytrace.py $NUM_NODES $NUM_WORKERS_PER_NODE 0.001 trivial $trace;
done)