Commit f70a0931 authored by Maiterth, Matthias's avatar Maiterth, Matthias
Browse files

Merge branch 'develop' into 'net-dev' 

# Conflicts:
#   raps/sim_config.py
#   raps/workloads/__init__.py
parents 0c1235b0 91c2da82

.gitignore

+2 −0
Original line number Diff line number Diff line
@@ -7,3 +7,5 @@ venv 
simulation_results/

models/fmu-models

.shell-completion-cache

raps-output-*

ppo_raps_logs

README.md

+8 −0
Original line number Diff line number Diff line
@@ -77,6 +77,14 @@ For MIT Supercloud 
    # Reinforcement learning test case

    raps train-rl --system mit_supercloud/part-cpu -f /opt/data/mit_supercloud/202201



Microsoft Azure - 2017 Philly Traces



    # Synthetic

    python main.py run-parts -x philly -w multitenant



    # Telemetry replay

    python main.py run-parts -x philly -f /opt/data/philly/trace-data --start 2017-10-03T00:14:56Z  --end 2017-10-04T00:00



For Lumi



    # Synthetic test for Lumi:

config/perlmutter.yaml

0 → 100644
+51 −0
Original line number Diff line number Diff line 
system:

  num_cdus: 36

  racks_per_cdu: 3

  nodes_per_rack: 128

  rectifiers_per_rack: 32

  chassis_per_rack: 8

  nodes_per_blade: 2

  switches_per_chassis: 4

  nics_per_node: 4

  rectifiers_per_chassis: 4

  nodes_per_rectifier: 4

  missing_racks: []

  down_nodes: []

  cpus_per_node: 1

  gpus_per_node: 4

  cpu_peak_flops: 3580000000000.0

  gpu_peak_flops: 9700000000000.0

  cpu_fp_ratio: 0.667

  gpu_fp_ratio: 0.667

power:

  power_gpu_idle: 88

  power_gpu_max: 300

  power_cpu_idle: 90

  power_cpu_max: 280

  power_mem: 74.26

  power_nic: 20

  power_nvme: 30

  power_switch: 250

  power_cdu: 8473.47

  power_update_freq: 15

  rectifier_peak_threshold: 13670

  sivoc_loss_constant: 13

  sivoc_efficiency: 0.98

  rectifier_loss_constant: 17

  rectifier_efficiency: 0.96

  power_cost: 0.094

scheduler:

  seed: 42

  job_arrival_time: 900

  mtbf: 11

  trace_quanta: 10

  min_wall_time: 3600

  max_wall_time: 43200

  ui_update_freq: 900

  max_nodes_per_job: 3000

  job_end_probs:

    COMPLETED: 0.63

    FAILED: 0.13

    CANCELLED: 0.12

    TIMEOUT: 0.11

    NODE_FAIL: 0.01

config/philly/2-gpu.yaml

0 → 100644
+51 −0
Original line number Diff line number Diff line 
system:

  num_cdus: 1

  racks_per_cdu: 1

  nodes_per_rack: 321

  chassis_per_rack: 3

  nodes_per_blade: 2

  switches_per_chassis: 4

  nics_per_node: 4

  rectifiers_per_chassis: 4

  nodes_per_rectifier: 4

  missing_racks: []

  down_nodes: []

  cpus_per_node: 2

  cores_per_cpu: 20

  gpus_per_node: 2

  cpu_peak_flops: 1248000000000.0 # assume Xeon E5-2690v4 CPU 64-bit

  gpu_peak_flops: 9300000000000.0 # assume 12G P100 32-bit

  cpu_fp_ratio: 0.667

  gpu_fp_ratio: 0.667

power:

  power_gpu_idle: 30

  power_gpu_max: 250

  power_cpu_idle: 90

  power_cpu_max: 270

  power_mem: 74.26

  power_nvme: 30

  power_nic: 20

  power_cdu: 8473.47

  power_switch: 250

  power_update_freq: 15

  rectifier_peak_threshold: 13670

  sivoc_loss_constant: 13

  sivoc_efficiency: 0.98

  rectifier_loss_constant: 17

  rectifier_efficiency: 0.96

  power_cost: 0.094

scheduler:

  multitenant: true

  job_arrival_time: 900

  mtbf: 11

  trace_quanta: 20

  min_wall_time: 3600

  max_wall_time: 43200

  ui_update_freq: 900

  max_nodes_per_job: 192

  job_end_probs:

    COMPLETED: 0.63

    FAILED: 0.13

    CANCELLED: 0.12

    TIMEOUT: 0.11

    NODE_FAIL: 0.01

config/philly/8-gpu.yaml

0 → 100644
+51 −0
Original line number Diff line number Diff line 
system:

  num_cdus: 1

  racks_per_cdu: 1

  nodes_per_rack: 231

  chassis_per_rack: 3

  nodes_per_blade: 2

  switches_per_chassis: 4

  nics_per_node: 4

  rectifiers_per_chassis: 4

  nodes_per_rectifier: 4

  missing_racks: []

  down_nodes: []

  cpus_per_node: 2

  cores_per_cpu: 20

  gpus_per_node: 8

  cpu_peak_flops: 1248000000000.0  # assume Xeon E5-2690v4 CPU 64-bit

  gpu_peak_flops: 12000000000000.0 # assume 24G P40 32-bit

  cpu_fp_ratio: 0.667

  gpu_fp_ratio: 0.667

power:

  power_gpu_idle: 50

  power_gpu_max: 250

  power_cpu_idle: 90

  power_cpu_max: 270

  power_mem: 74.26

  power_nvme: 30

  power_nic: 20

  power_cdu: 8473.47

  power_switch: 250

  power_update_freq: 15

  rectifier_peak_threshold: 13670

  sivoc_loss_constant: 13

  sivoc_efficiency: 0.98

  rectifier_loss_constant: 17

  rectifier_efficiency: 0.96

  power_cost: 0.094

scheduler:

  multitenant: true

  job_arrival_time: 900

  mtbf: 11

  trace_quanta: 20

  min_wall_time: 3600

  max_wall_time: 43200

  ui_update_freq: 900

  max_nodes_per_job: 192

  job_end_probs:

    COMPLETED: 0.63

    FAILED: 0.13

    CANCELLED: 0.12

    TIMEOUT: 0.11

    NODE_FAIL: 0.01