Benchmarks

Docker vs Standard Linux

I wanted to benchmark running srim calculations on normal linux vs within a docker container. TRIM on linux showed more variability between job runs. When I refer to TRIM on linux it is equivalent to wine TRIM.exe.

I used the parallel command to test differing number of cores. Yes this is a simple approach but will be in the right ballpark and trends are certainly compelling for the docker container. All test we done on my two core Lenovo t440s.

Docker startup cost is 8.5 seconds vs 5.8 seconds for standard linux process. But the docker container is significantly faster using =xvfb-run= best performance is linux 8.3 ions/second vs docker 13.2 ions/second. These tests indicate that the docker container may have better performance due to rending in a virtual X frame buffer.

The python simple script to be run is a Nickel in Nickel irradiation.

import os
from srim import Ion, Layer, Target, TRIM

ion = Ion('Ni', energy=3.0e6)
layer = Layer({
        'Ni': {
            'stoich': 1.0,
            'E_d': 30.0,
            'lattice': 0.0,
            'surface': 3.0
        }}, density=8.9, width=20000.0)
target = Target([layer])
trim = TRIM(target, ion, number_ions=100, calculation=1)
srim_executable_directory = '/tmp/srim'
results = trim.run(srim_executable_directory)
os.makedirs('/tmp/output', exist_ok=True)
TRIM.copy_output_files('/tmp/srim', '/tmp/output')

And the benchmarks results.

time parallel -j 6 python ni.py -- 1 2

ions	cores	linux [s]	linux [ion/s]	linux [ion/s core]
1	1	5.8	0.17241379	0.17241379
100	1	29	3.4482759	3.4482759
200	1	53	3.7735849	3.7735849
200	2	35	5.7142857	2.8571429
400	2	63	6.3492063	3.1746032
300	3	43	6.9767442	2.3255814
600	3	79	7.5949367	2.5316456
400	4	55	7.2727273	1.8181818
800	4	101	7.9207921	1.9801980
1600	4	192	8.3333333	2.0833333
500	5	65	7.6923077	1.5384615
1000	5	121	8.2644628	1.6528926

time parallel -j 6 \
         docker run \
              -v $PWD/examples/docker/:/opt/pysrim/ \
              -v /tmp/output:/tmp/output
              -it costrouc/pysrim sh -c "xvfb-run -a python3.6 /opt/pysrim/ni.py" -- 1 2

ions	cores	docker [s]	docker [ion/s]	docker [ion/s core]
1	1	8.5	0.11764706	0.11764706
100	1	27	3.7037037	3.7037037
200	1	46	4.3478261	4.3478261
200	2	31	6.4516129	3.2258065
400	2	52	7.6923077	3.8461538
300	3	39	7.6923077	2.5641026
600	3	69	8.6956522	2.8985507
400	4	39	10.256410	2.5641026
800	4	67	11.940299	2.9850746
1600	4	121	13.223140	3.3057851
500	5	51	9.8039216	1.9607843
1000	5	69	14.492754	2.8985507