A process to submit an amber job to a cluster for a range of Processors.
The metrics are gathered to record the results in ns/day to see the scaling metrics.
Requirements:
-
R software = R is needed to generate the png graph of the results. Obtained from https://www.r-project.org/
-
Amber software = Amber is purchased from http://ambermd.org
-
A web server that has User_dir enabled for Apache. This allows users to have a website at http://web-server/~username that is located at /home/username/public_html.
-
Torque PBS queue server is assumed for job creation.
Step 1: clone the repository with git.
% git clone https://github.com/hamhpc/amber_cluster_benchmark.git
% cd amber_cluster_benchmark
Step 2: edit the configuration.
Edit the run_tests.sh script to make sure that it will operate in your environment.
% vi run_tests.sh
Step 2a: add your input files.
You can add your input files into the etc directory of the amber_cluster_benchmark directory.
Be sure to update the names of the files in the step 2 (above).
Step 3: run the tests.
% ./run_tests.sh
This will submit a bunch of jobs to the queue. It's a sequence of NPROC's from 2 up to the max.
You can specify the NCPUS in the run_test.sh script. The jobs will take some time.
The large ones will complete first and the last to complete is the 2 CPU.
It usually finishes in under 1.5 hours which is what the walltime is set for the job runs.
Step 4: compile the results.
After all the jobs are run and off the queue you'll need to gather the results and create the web page and graph.
% cd ~/amber_cluster_bechmark/results_Jan-06-2017-15:27 (note it'll have the date of when you ran these tests)
% ./make_web.sh
This will run the R script to generate the graph called make_graph.R.
It'll also build the results in your public_html directory.
From the head node webserver access:
http://<head_node_server>/~<YOUR USERNAME>/amber_cluster_benchmark/index.html
Step 5 (optional): Get GPU results.
After building the cluster page and graph now it's time to run the gpu tests and have that image added to the index.html.
% cd ~/amber_cluster_benchmark/
% ./gpu_tests.sh
This will run the gpu tests. Make sure to edit the gpu_tests.sh script to suit your environment.
Step 6 (optional): compile the GPU results.
After all the jobs are run and off the queue you'll need to gather the results and create the GPU graph.
% cd ~/amber_cluster_bechmark/results_gpu-Jan-06-2017-15:27 (note it'll have the date of when you ran these tests)
Before making the graph, edit the results file (results-gpu.csv) and change the hosts to reflect the short_hostname:#.
This is so the graph will look and fit better in the image.
Change gpunode.domain.edu/1 to gpunode:1 to represent the second card on that host.
Now build the graph image by running:
% ./make_web.sh
This will run the R script to generate the graph called make_graph.R.
It'll also build the results in your public_html directory.
From the head node webserver access:
http://<head_node_server>/~<YOUR USERNAME>/amber_cluster_benchmark/index.html
You should now see the GPU graph on the page in addition to the previous cluster test.