Name: sc-benchmark
Owner: Fred Hutchinson Cancer Research Center
Description: Scientific Computing benchmarks (HPC, storage, genomics)
Created: 2014-08-13 00:00:26.0
Updated: 2017-05-26 19:04:49.0
Pushed: 2018-01-06 19:51:27.0
Homepage: null
Size: 940
Language: Shell
GitHub Committers
| User | Most Recent Commit | # Commits |
|---|---|---|
| Jeffrey Katcher | 2017-08-21 20:01:06.0 | 5 |
| Dirk Petersen | 2018-01-06 19:51:22.0 | 20 |
| Robert McDermott | 2017-06-21 17:55:30.0 | 6 |
Other Committers
| User | Most Recent Commit | # Commits | |
|---|---|---|---|
| root | root@gizmog10.fhcrc.org | 2015-12-13 17:04:20.0 | 1 |
Scientific Computing benchmarks (HPC, storage, genomics)
using a slightly modified version of this code : https://www.rdocumentation.org/packages/rbenchmark/versions/1.0.0/topics/benchmark
We ran tests using R 3.3.2 on Ubuntu 14.04 after AWS had patched for the meltdown vulnerability. 22 sec local vs 25 sec on AWS suggests the impact of the meltdown patches on CPU intensive workflows such as statistical simulations is relatively small. The AWS instance has ca 7% better Passmark benchmarks (measured before the patching) despite clocking a bit lower according to https://www.cpubenchmark.net/. This suggests that AWS is overclocking their CPUs.
test replications elapsed relative user.self sys.self user.child sys.child
1 1:10^6 100 0.106 1 0.074 0.032 0 0
test elapsed replications
1 random.array(100, 100) 0.811 1000
2 random.array(100, 100) 0.805 1000
3 random.array(100, 100) 0.805 1000
4 random.array(100, 100) 0.817 1000
5 random.array(100, 100) 0.808 1000
6 random.array(100, 100) 0.805 1000
test replications elapsed average
2 arr 10 0.008 0.000800
4 arr 100 0.080 0.000800
6 arr 1000 0.797 0.000797
1 rep 10 0.085 0.008500
3 rep 100 0.846 0.008460
5 rep 1000 8.601 0.008601
test elapsed replications
2 arr 0.803 1000
1 rep 8.585 1000
real 0m25.01s
user 0m0.78s
sys 0m0.32s
$ time ./rbenchmark.R
test replications elapsed relative user.self sys.self user.child sys.child
1 1:10^6 100 0.133 1 0.044 0.089 0 0
test elapsed replications
1 random.array(100, 100) 0.780 1000
2 random.array(100, 100) 0.685 1000
3 random.array(100, 100) 0.687 1000
4 random.array(100, 100) 0.695 1000
5 random.array(100, 100) 0.685 1000
6 random.array(100, 100) 0.687 1000
test replications elapsed average
2 arr 10 0.007 0.000700
4 arr 100 0.069 0.000690
6 arr 1000 0.677 0.000677
1 rep 10 0.074 0.007400
3 rep 100 0.729 0.007290
5 rep 1000 7.320 0.007320
test elapsed replications
2 arr 0.683 1000
1 rep 7.294 1000
real 0m21.827s
user 0m21.319s
sys 0m0.314s
user@box:$ ~/py/scratch-dna.py
Error! 4 arguments required.
Execute scratch-dna.py [number-of-files] [file-size-bytes] [random-file-size-multiplier] [writable-directory]
scratch-dna.py is a storage benchmarking tool that creates a random DNA string, concatenates this string a random number of times and writes the result multiple files with varying length.
2 examples:
user@box:$ ~/py/scratch-dna.py 1000 2000 3000 .
box: ... building random DNA sequence of 1.953 KB...
box: ... writing 1000 files with filesizes between 1.953 KB and 5,859.375 KB ...
box: 2897.315644 MB written in 8.609 seconds (336.559 MB/s, 116 files/s)
here we are writing 1000 files into the current directory to evaluate throughput performance. The files contain a dna snippet that is 2000 bytes long and each snippet is contatenated with the running file counter and the current hostname and then duplicated n times (n is a random number between 1 and 3000).
user@box:$ ~/py/scratch-dna.py 10000 4 1 .
box: ... building random DNA sequence of 0.004 KB...
box: ... writing 10000 files with filesizes between 0.004 KB and 0.004 KB ...
box: 0.141993 MB written in 20.721 seconds (0.007 MB/s, 482 files/s)
box:$ cat box/scr-box-file-6985-1
6985CATCbox
here we are writing 10000 small files (max 12 bytes) into the current directory to evaluate metadata performance. The dna snippet is 4 bytes long and is not duplicated.
In general dna snippet would be much more random if it could be regenerated for each file, however this would increase compute times. If we run this scratch-dna.py on a compute cluster each node would generate unique data.
A drop in replacement for scratch-dna.py written in Go that adds the ability to specify how many files to write in parallel (-p num). Also adds verbose output (-v) that shows progress as files are being written.
ratch-dna-go 1000 1048576 10 ~/dna
ding random DNA sequence of 1.0 MB...
ing 1000 files with filesizes between 1.0 MB and 10.0 MB...
!
er of Files Written: 1000, Total Size: 5.4GiB, Avg FPS: 34, Avg Throughput: 190 MiB/s, Elapsed Time: 29 seconds
ratch-dna-go -v -p 8 1000 1048576 10 ~/dna
ding random DNA sequence of 1.0 MB...
ing 1000 files with filesizes between 1.0 MB and 10.0 MB...
s Completed: 139, Data Written: 0.7GiB, Files Remaining: 868, Cur FPS: 139, Throughput: 745 MiB/s
s Completed: 277, Data Written: 1.4GiB, Files Remaining: 730, Cur FPS: 138, Throughput: 738 MiB/s
s Completed: 354, Data Written: 1.8GiB, Files Remaining: 652, Cur FPS: 118, Throughput: 626 MiB/s
s Completed: 370, Data Written: 1.9GiB, Files Remaining: 635, Cur FPS: 92, Throughput: 494 MiB/s
s Completed: 428, Data Written: 2.2GiB, Files Remaining: 578, Cur FPS: 85, Throughput: 460 MiB/s
s Completed: 529, Data Written: 2.8GiB, Files Remaining: 477, Cur FPS: 88, Throughput: 479 MiB/s
s Completed: 672, Data Written: 3.5GiB, Files Remaining: 332, Cur FPS: 96, Throughput: 517 MiB/s
s Completed: 805, Data Written: 4.3GiB, Files Remaining: 201, Cur FPS: 100, Throughput: 545 MiB/s
s Completed: 938, Data Written: 4.9GiB, Files Remaining: 67, Cur FPS: 104, Throughput: 562 MiB/s
!
er of Files Written: 1000, Total Size: 5.3GiB, Avg FPS: 100, Avg Throughput: 541 MiB/s, Elapsed Time: 10 seconds