Commit a7515fb3 authored by Maiterth, Matthias's avatar Maiterth, Matthias
Browse files

Merge branch 'rl3' into 'develop' 

RL framework (try 3)

See merge request !117
parents 15aacee4 c41b265e

README.md

+8 −3
Original line number Diff line number Diff line
@@ -62,16 +62,21 @@ For MIT Supercloud 
    python -m raps.dataloaders.mit_supercloud.cli download --start 2021-05-21T13:00 --end 2021-05-21T14:00



    # Load data and run simulation - will save data as part-cpu.npz and part-gpu.npz files

    raps run-parts -x mit_supercloud -f $DPATH --system mit_supercloud --start 2021-05-21T13:00 --end 2021-05-21T14:00

    raps run-parts -x mit_supercloud -f $DPATH --start 2021-05-21T13:00 --end 2021-05-21T14:00

    # or simply

    raps run-parts experiments/mit-replay-25hrs.yaml

    # Note: if no start, end dates provided will default to run 24 hours between

    # 2021-05-21T00:00 to 2021-05-22T00:00 set by defaults in raps/dataloaders/mit_supercloud/utils.py



    # Re-run simulation using npz files (much faster load)

    raps run-parts -x mit_supercloud -f part-*.npz --system mit_supercloud

    raps run-parts -x mit_supercloud -f part-*.npz



    # Synthetic tests for verification studies:

    raps run-parts -x mit_supercloud -w multitenant



    # Reinforcement learning test case

    python main.py train-rl --system mit_supercloud/part-cpu -f /opt/data/mit_supercloud/202201



For Lumi



    # Synthetic test for Lumi:

config/mit_supercloud/part-cpu.yaml

+1 −1
Original line number Diff line number Diff line
@@ -36,7 +36,7 @@ power: 
  power_cost: 0.094

scheduler:

  multitenant: true

  job_arrival_time: 900

  job_arrival_time: 1

  mtbf: 11

  trace_quanta: 10

  min_wall_time: 3600

experiments/frontier.yaml

+2 −2
Original line number Diff line number Diff line 
system: frontier

replay:

  - ~/data/frontier/slurm/joblive/date=2024-01-18

  - ~/data/frontier/jobprofile/date=2024-01-18
 
  - /opt/data/frontier/slurm/joblive/date=2024-01-18

  - /opt/data/frontier/jobprofile/date=2024-01-18

experiments/gcloudv2.yaml

+1 −1
Original line number Diff line number Diff line 
system: gcloudv2

replay:

  - ~/data/gcloud/v2/google_cluster_data_2011_sample

  - /opt/data/gcloud/v2/google_cluster_data_2011_sample

ff: 600

experiments/marconi100.yaml

+1 −1
Original line number Diff line number Diff line 
system: marconi100

replay:

  - ~/data/marconi100/job_table.parquet
 
  - /opt/data/marconi100/job_table.parquet