[Radical-Cylon] Summit build and test

rivanna/scripts/rp-experiment-setup.py

@@ -85,7 +85,7 @@
   echo "..............................................................." 
   export RADICAL_LOG_LVL="DEBUG"
   export RADICAL_PROFILE="TRUE"
-  export RADICAL_PILOT_DBURL="mongodb://rct-tutorial:[email protected]:27017/rct-tutorial"
+  export RADICAL_PILOT_DBURL="mongodb://Your_mongo_db_url"
   echo "..............................................................."  
   lscpu
   echo "..............................................................."

summit/README.md

@@ -1,6 +1,6 @@
 # Running Cylon on Summit
 
-Gregor von Laszewski ([email protected]), and Niranda Perera
+Gregor von Laszewski ([email protected]), Niranda Perera, and Arup Sarker([email protected], [email protected])
 
 ## Issues on Summit
 
@@ -51,7 +51,8 @@ cd cylon
 git checkout summit
 cd ~
 
-module load python/3.7.7    
+module load python/3.7.7
+module load gcc/9.3.0 
 python -m venv $HOME/CYLON
 source $HOME/CYLON/bin/activate
 
@@ -65,7 +66,6 @@ pip install -U pytest-mpi
 pip install cmake
 pip install numpy
 
-module load gcc/9.3.0
 
 cd ~/cylon
 rm -rf build
@@ -161,4 +161,4 @@ A script that changes the parameter of n is located in the scripts dir and can b
 
 ```bash
 sh ./benchmark-summit.sh 2>&1 | tee -a summit.log
-```
+```

summit/rp/Makefile

@@ -0,0 +1,42 @@
+SHELL=/bin/bash
+
+.PHONY: load image-singularity image-docker project
+
+all: ${EXECS}
+
+login:
+	ssh -tt rivanna "/opt/rci/bin/ijob --partition=parallel --account=bii_dsc_community --time=30:00 --ntasks-per-node=4 --nodes=2"
+
+load:
+	./load.sh
+
+clean:
+	rm -f *.log *.err script-*.slurm cylonrun-* script-*.sh script-*.lsf
+	rm -r raptor-*.cfg rp.session.*
+
+rp: 
+	python rp-experiment-setup.py
+
+cy: load
+	python cylon-experiment-setup.py
+
+q:
+	bjobs --format="%.18i %.9P %.50j %.8u %.8T %.10M %.9l %.6D %R" --me
+
+a:
+	bjobs --format="%all" --me
+
+
+qq:
+	watch bjobs -u arupcseuva --format=\"%.18i %.9P %.50j %.8u %.8T %.10M %.9l %.6D %R\"
+
+i:
+	cat out.log
+	cat out.err
+	fgrep "###"  out.log | wc -l
+
+
+cancel:
+	- ./cancel.sh
+	- bjobs -u ${USER} 
+

summit/rp/README.md

@@ -0,0 +1,40 @@
+1. Load module and activate the python virtual environment
+
+```
+module load gcc/9.2.0 openmpi/3.1.6 cmake/3.23.3 python/3.7.7 
+source /path_to_virtual_environment/cylon_rp_venv/bin/activate
+```
+
+2. Install Cloudmesh and Radical-Pilot
+
+```
+pip install cloudmesh-common
+pip install openssl-python
+python3 -m pip install urllib3==1.26.6
+pip install radical.pilot
+```
+
+3. Make change in the ```cylon-experiment-setup.py ``` or ```rp-experiment-setup.py ``` for the configurations changes.
+
+```
+combination = [\
+    # (1,4, 5000, "parallel", "exclusive"), # always pending
+    (2,37, 1000000, "parallel", ""), 
+    (4,37, 35000000, "parallel", ""), 
+    (6,37, 35000000, "parallel", ""), 
+    (8,37, 35000000, "parallel", ""), 
+    (10,37, 35000000, "parallel", ""), 
+    (12,37, 35000000, "parallel", ""), 
+    (14,37, 35000000, "parallel", ""),
+]
+```
+
+
+4. Run the scripts  as follows.
+
+```bash
+make clean # For cleaning 
+make rp # For radical pilot
+make cy # for bear metal Cylon
+
+```

summit/rp/bkp_script-02-42.sh

@@ -0,0 +1,36 @@
+#!/bin/bash
+
+##BSUB -P gen150_bench
+##BSUB -W 0:45
+##BSUB -nnodes 2
+##BSUB -alloc_flags smt1
+##BSUB -J cylonrun-w-2
+##BSUB -o cylonrun-w-2.%J
+##BSUB -e cylonrun-w-2.%J
+
+#module load python/3.7.7
+#source $HOME/CYLON/bin/activate
+
+#module load gcc/9.3.0
+echo "..............................................................." 
+export RADICAL_LOG_LVL="DEBUG"
+export RADICAL_PROFILE="TRUE"
+export RADICAL_PILOT_DBURL="mongodb://You_Mongodb_url"
+echo "..............................................................."  
+lscpu
+echo "..............................................................."
+BUILD_PATH=$HOME/project/dev/cylon/build
+export LD_LIBRARY_PATH=$BUILD_PATH/arrow/install/lib64:$BUILD_PATH/glog/install/lib64:$BUILD_PATH/lib64:$BUILD_PATH/lib:$LD_LIBRARY_PATH
+
+
+echo ################## 2 case
+time  python $HOME/project/dev/cylon/summit/rp/rp_scaling.py raptor-2-42.cfg 
+#-n 50000000 -s w
+
+
+if ((0)); then
+   time  python $HOME/project/dev/cylon/summit/rp/rp_scaling.py raptor-2-42.cfg 
+   #-n 50000000 -s w
+fi
+#time python rp_scaling.py raptor-2-42.cfg
+echo "..............................................................."

summit/rp/cancel.sh

@@ -0,0 +1,2 @@
+#! /bin/sh
+squeue -u $USER | awk '{print $1}' | tail -n+2 | xargs scancel

summit/rp/cylon-experiment-setup.py

@@ -0,0 +1,117 @@
+import os
+import sys
+from textwrap import dedent
+from cloudmesh.common.util import writefile
+from cloudmesh.common.util import readfile
+from cloudmesh.common.util import banner
+from cloudmesh.common.console import Console
+
+counter = 0
+
+debug = True
+debug = False
+
+partition="bii-gpu"
+
+partition="parallel"
+
+
+# (nodes, threads, rows, partition, "exclusive")
+combination = [\
+    # (1,4, 5000, "parallel", "exclusive"), # always pending
+    #(2,37, 1000000, "parallel", ""), 
+    (4,37, 35000000, "parallel", ""), 
+    #(6,37, 35000000, "parallel", ""), 
+    #(8,37, 35000000, "parallel", ""), 
+    #(10,37, 35000000, "parallel", ""), 
+    #(12,37, 35000000, "parallel", ""), 
+    #(14,37, 35000000, "parallel", ""),
+]
+
+'''
+combination = []
+for nodes in range(0,50):
+  for threads in range(0,37):
+    combination.append((nodes+1, threads+1, "parallel", "")) 
+'''
+
+total = len(combination)
+jobid="-%j"
+# jobid=""
+
+f = open("submit.log", "w")
+for nodes, threads, rows, partition, exclusive in combination:
+  counter = counter + 1
+
+  if exclusive == "exclusive":
+      exclusive = "#SBATCH --exclusive"
+      e = "e1"
+  else:
+      exclusive = ""
+      e = "e0"
+
+  usable_threads = nodes * threads
+
+  '''
+  cores_per_node = nodes * threads - 2
+
+  print (cores_per_node)
+
+  config = readfile("raptor.in.cfg")
+
+  config = config.replace("CORES_PER_NODE", str(cores_per_node))
+  config = config.replace("NO_OF_ROWS", str(rows))
+
+
+  print (config)
+
+  cfg_filename = f"raptor-{nodes}-{threads}.cfg"
+
+  writefile(cfg_filename, config)
+  '''
+  banner(f"SLURM {nodes} {threads} {counter}/{total}")
+  script=dedent(f"""
+  #!/bin/bash
+  #SBATCH --job-name=h-n={nodes:02d}-t={threads:02d}-e={e}
+  #SBATCH --nodes={nodes}
+  #SBATCH --ntasks-per-node={threads}
+  #SBATCH --time=15:00
+  #SBATCH --output=out-{nodes:02d}-{threads:02d}{jobid}.log
+  #SBATCH --error=out-{nodes:02d}-{threads:02d}{jobid}.err
+  #SBATCH --partition=parallel
+  #SBATCH -A bii_dsc_community
+  {exclusive}
+  echo "..............................................................."
+  module load gcc/9.2.0 openmpi/3.1.6 cmake/3.23.3 python/3.7.7
+  echo "..............................................................."
+  source /project/bii_dsc_community/djy8hg/cylon_rp_venv/bin/activate
+  echo "..............................................................." 
+  BUILD_PATH=/project/bii_dsc_community/djy8hg/Project/cylon/build
+  echo "..............................................................."  
+  export LD_LIBRARY_PATH=$BUILD_PATH/arrow/install/lib64:$BUILD_PATH/glog/install/lib64:$BUILD_PATH/lib64:$BUILD_PATH/lib:$LD_LIBRARY_PATH  
+  echo "..............................................................."  
+  which python gcc g++
+  echo "..............................................................."  
+  lscpu
+  echo "..............................................................."
+  time srun --exact --nodes {nodes} --ntasks {usable_threads}  python cylon_scaling.py -n {rows}
+  echo "..............................................................."
+  """).strip()
+
+  print (script)
+  filename = f"script-{nodes:02d}-{threads:02d}.slurm"
+  writefile(filename, script)
+
+
+  if not debug:
+
+    r = os.system(f"sbatch {filename}")
+    total = nodes * threads
+    if r == 0:
+      msg = f"{counter} submitted: nodes={nodes:02d} threads={threads:02d} total={total}"
+      Console.ok(msg)
+    else:
+      msg = f"{counter} failed: nodes={nodes:02d} threads={threads:02d} total={total}"
+      Console.error(msg)
+    f.writelines([msg, "\n"])
+f.close()