From b95c31d5c6ac5f823a28f4bc4ba987cdd1402d84 Mon Sep 17 00:00:00 2001 From: DIR-LAB <45381188+DIR-LAB@users.noreply.github.com> Date: Fri, 29 May 2020 14:39:09 -0400 Subject: [PATCH] Update README.md --- README.md | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 3cce2e0..fd87e3b 100644 --- a/README.md +++ b/README.md @@ -84,7 +84,7 @@ There are many parameters you can use: * `--backfil`, enable/disable backfilling during the test * `--score_type`, specify the scheduling metrics. [0]：bounded job slowdown；[1]: job waiting time; [2]: job response time; [3] system resource utilization. -## Step-By-Step Example +## A Step-By-Step Example Here, we give a step-by-step example to show the complete training/monitoring/testing workflow of RLScheduler. @@ -92,6 +92,7 @@ Here, we give a step-by-step example to show the complete training/monitoring/te ```bash python ppo-pick-jobs.py --workload "./data/lublin_256.swf" --exp_name lublin256-seed0 --trajs 500 --seed 0 ``` +In this experiment, we have `seed=0`, collect 500 trajectories in each epoch, and optimize average bounded slowdown. * Step 2: Monitor the training by checking the training curves ```bash @@ -106,8 +107,8 @@ It will output something like this: ```bash python compare-pick-jobs.py --rlmodel "./data/logs/lublin256-seed0/lublin256-seed0_s0/" --workload "./data/lublin_256.swf --seed 1 --len 1024 --iter 10" ``` -It will output something like this: +In this scheduling case, we randomly select 10 job sequences using `seed=1`. It will output something like this for comparing different scheduling results:
Lublin256 Training Curve
- +We use the average to produce the performance tables in the paper.