Batsim performance test on HPC2N and RICC

default.nix

{
  pkgs ? import (
    fetchTarball "https://github.com/NixOS/nixpkgs/archive/17.09.tar.gz") {},
  dmpkgs ? import (
    fetchTarball "https://gitlab.inria.fr/vreis/datamove-nix/repository/7e2af348a5993de734452cd1cd8e8fafd76e308c/archive.tar.gz"
    ) { inherit pkgs;},
}:

pkgs.stdenv.mkDerivation rec {
  name = "expe-env";
  env = pkgs.buildEnv { name = name; paths = buildInputs; };
  buildInputs = [
    pkgs.curl
    pkgs.gzip
    pkgs.python35
    pkgs.python35Packages.sortedcontainers
    dmpkgs.batsim
    dmpkgs.batsched
    dmpkgs.batexpe
    pkgs.R
    pkgs.rPackages.dplyr
    pkgs.rPackages.readr
  ];
}

execution.log

# make run
nix-shell --pure ./default.nix --command ./run_instances.bash
downloading ‘https://gitlab.inria.fr/vreis/datamove-nix/repository/7e2af348a5993de734452cd1cd8e8fafd76e308c/archive.tar.gz’... [0/0 KiB, 0.0 KiB/s]
+ robin --quiet ./instances/hpc2n-fcfs.robin

real	1m27.591s
user	1m4.549s
sys	0m19.965s
+ robin --quiet ./instances/hpc2n-easy.robin

real	1m24.661s
user	1m4.405s
sys	0m18.813s
+ robin --quiet ./instances/ricc-fcfs.robin

real	12m42.988s
user	11m36.551s
sys	0m51.017s
+ robin --quiet ./instances/ricc-easy.robin

real	14m57.913s
user	13m40.592s
sys	0m54.568s

Makefile

default:
	@echo "There is no default target. Try make prepare, make run or make analyze."

prepare:
	nix-shell --pure ./default.nix --command ./prepare_instances.bash

run:
	nix-shell --pure ./default.nix --command ./run_instances.bash

analyze:
	nix-shell --pure ./default.nix --command ./run_analysis.R

clean:
	\rm -rf instances/ platforms/ workloads/

mrproper: clean
	\rm -rf result/

distclean: mrproper

merged_result.csv

prepare_instances.bash

#!/usr/bin/env bash
set -eux

#############
# Variables #
#############
# A clone to Batsim's repo, on tag v2.0.0
BATSIM_DIR=$(realpath ./batsim-v2.0.0)
RESULT_DIR="$(realpath .)/result"
BATLOG="--quiet --disable-schedule-tracing --disable-machine-state-tracing"
TIMEOUTS="--simulation-timeout=86400 --ready-timeout=5 --success-timeout=120 --failure-timeout=0"

######################
# Generate platforms #
######################
mkdir -p platforms
${BATSIM_DIR}/tools/energy_pform_generator_homo_nonet.py -n 240 \
    -o platforms/platform_hpc2n.xml
${BATSIM_DIR}/tools/energy_pform_generator_homo_nonet.py -n 8192 \
    -o platforms/platform_ricc.xml

######################
# Generate workloads #
######################
mkdir -p workloads
curl -o ./workloads/HPC2N-2002-2.2-cln.swf.gz \
    http://www.cs.huji.ac.il/labs/parallel/workload/l_hpc2n/HPC2N-2002-2.2-cln.swf.gz
gunzip -k -f ./workloads/HPC2N-2002-2.2-cln.swf.gz
${BATSIM_DIR}/tools/swf_to_batsim_workload_delay.py \
    ./workloads/HPC2N-2002-2.2-cln.swf ./workloads/workload_hpc2n.json \
    -i1 -t -pf 240

curl -o ./workloads/RICC-2010-2.swf.gz \
    http://www.cs.huji.ac.il/labs/parallel/workload/l_ricc/RICC-2010-2.swf.gz
gunzip -k -f ./workloads/RICC-2010-2.swf.gz
${BATSIM_DIR}/tools/swf_to_batsim_workload_delay.py \
    ./workloads/RICC-2010-2.swf ./workloads/workload_ricc.json \
    -i1 -t -pf 8192

###############################
# Define simulation instances #
###############################
mkdir -p instances

# hpc2n-fcfs
robin generate ./instances/hpc2n-fcfs.robin \
    --output-dir="${RESULT_DIR}/hpc2n-fcfs" \
    --batcmd="batsim ${BATLOG} -p $(realpath ./platforms/platform_hpc2n.xml) -w $(realpath ./workloads/workload_hpc2n.json) -e ${RESULT_DIR}/hpc2n-fcfs/out" \
    --schedcmd='batsched -v fcfs_fast >/dev/null' \
    ${TIMEOUTS}

# ricc-fcfs
robin generate ./instances/ricc-fcfs.robin \
    --output-dir="${RESULT_DIR}/ricc-fcfs" \
    --batcmd="batsim ${BATLOG} -p $(realpath ./platforms/platform_ricc.xml) -w $(realpath ./workloads/workload_ricc.json) -e ${RESULT_DIR}/ricc-fcfs/out" \
    --schedcmd='batsched -v fcfs_fast >/dev/null' \
    ${TIMEOUTS}

# hpc2n-easy
robin generate ./instances/hpc2n-easy.robin \
    --output-dir="${RESULT_DIR}/hpc2n-easy" \
    --batcmd="batsim ${BATLOG} -p $(realpath ./platforms/platform_hpc2n.xml) -w $(realpath ./workloads/workload_hpc2n.json) -e ${RESULT_DIR}/hpc2n-easy/out" \
    --schedcmd='batsched -v easy_bf_fast >/dev/null' \
    ${TIMEOUTS}

# ricc-easy
robin generate ./instances/ricc-easy.robin \
    --output-dir="${RESULT_DIR}/ricc-easy" \
    --batcmd="batsim ${BATLOG} -p $(realpath ./platforms/platform_ricc.xml) -w $(realpath ./workloads/workload_ricc.json) -e ${RESULT_DIR}/ricc-easy/out" \
    --schedcmd='batsched -v easy_bf_fast >/dev/null' \
    ${TIMEOUTS}

reproduce.bash

#!/usr/bin/env bash
# This file describes how to reproduce the whole experiment

# Step 0 (retrieve gist)
curl -o ./gist.zip https://codeload.github.com/gist/56ec34c71fbe73bc8fb6ead959bdc2e9/zip/master
unzip ./gist.zip
mv 56ec34c71fbe73bc8fb6ead959bdc2e9-master ./gist
cd ./gist
chmod +x ./*.bash ./*.R

# Step 1 (install Nix if needed)
curl https://nixos.org/nix/install | sh

# Step 2 (retrieve a copy of Batsim repo at version v2.0.0)
BATSIM_DIR=./batsim-v2.0.0
git clone https://gitlab.inria.fr/batsim/batsim.git ${BATSIM_DIR}
pushd .
cd ${BATSIM_DIR}
git checkout v2.0.0
popd

# Step 3 (update BATSIM_DIR in prepare_instances.bash if needed)

# Step 4 (prepare the instances)
# First run may be very long, as Nix will retrieve ALL needed dependencies.
make prepare

# Step 5 (run the instances)
make run

# Step 6 (analyze results)
make analyze

run_analysis.R

#!/usr/bin/env Rscript
library("readr")
library("dplyr")

hpc2n_easy = read_csv("result/hpc2n-easy/out_schedule.csv") %>%
    mutate(workload='hpc2n', algo='easy')
hpc2n_fcfs = read_csv("result/hpc2n-fcfs/out_schedule.csv") %>%
    mutate(workload='hpc2n', algo='fcfs')
ricc_easy = read_csv("result/ricc-easy/out_schedule.csv") %>%
    mutate(workload='ricc', algo='easy')
ricc_fcfs = read_csv("result/ricc-fcfs/out_schedule.csv") %>%
    mutate(workload='ricc', algo='fcfs')

merged = bind_rows(hpc2n_easy, hpc2n_fcfs, ricc_easy, ricc_fcfs) %>%
    select(workload, algo, scheduling_time, simulation_time) %>%
    mutate(scheduling_time_proportion= scheduling_time / simulation_time) %>%
    arrange(workload, algo)
print(merged)
write_csv(merged, "result/merged_result.csv")

run_instances.bash

#!/usr/bin/env bash
set -eux

################################
# Execute simulation instances #
################################
time robin --quiet ./instances/hpc2n-fcfs.robin
time robin --quiet ./instances/hpc2n-easy.robin
time robin --quiet ./instances/ricc-fcfs.robin
time robin --quiet ./instances/ricc-easy.robin