[test] Add allocation speed check #597
Conversation
victorusu
changed the title
|
@toxa81, can you help me with your thoughts on the test? I am not sure why I am getting a huge fluctuation in both checks. The worst results of huge pages are consistently 10 times smaller than the worst results of no hugepages. But both fluctuate by 6 orders of magnitude. |
victorusu
force-pushed
the
test/alloc_speed
branch
from
c4b6dfd to
7b32d6e
Compare
The test is inconclusive. I am still making the PR in order to discuss possible solutions. The allocation times fluctuate from 1.90735e-06 to ~2.5 sec. I am not sure if this is because of the OS memory load by the time of the execution of if this is related to the small memory allocation of the test.
victorusu
force-pushed
the
test/alloc_speed
branch
from
7b32d6e to
fee7c73
Compare
toxa81
left a comment
toxa81
left a comment
There was a problem hiding this comment.
I post a modification line in the code review to avoid potential over-optimisations. Regarding the fluctuations: I don't have an explanation, but the time 1.90735e-06 sec. seems too short. Somebody familiar with Linux architecture may give an advise.
Codecov Report
@@ Coverage Diff @@
## master #597 +/- ##
=======================================
Coverage 91.52% 91.52%
=======================================
Files 72 72
Lines 9142 9142
=======================================
Hits 8367 8367
Misses 775 775Continue to review full report at Codecov.
|
teojgo
changed the title
vkarak
changed the title
vkarak
left a comment
vkarak
left a comment
There was a problem hiding this comment.
Apart from my specific comments, this PR should fix also the following two problems:
- The test must be placed under
microbenchmarks - The reference values seem completely wrong. Please verify.
2018-12-17T16:48:58|reframe 2.16-dev0|AllocSpeedTest_2M on dom:gpu using PrgEnv-gnu|jobid=906488|perf=0.106014|ref=0.16 (l=None, u=0.1)|None
2018-12-17T16:49:01|reframe 2.16-dev0|AllocSpeedTest_2M on dom:mc using PrgEnv-gnu|jobid=906489|perf=0.189877|ref=0.5 (l=None, u=0.1)|None
- Please use units in the reference tuple.
|
@jenkins-cscs retry kesch |
6 participants
The test is inconclusive. I am still making the PR in order to discuss
possible solutions.
The allocation times fluctuate from 1.90735e-06 to ~2.5 sec.
I am not sure if this is because of the OS memory load by the time of
the execution of if this is related to the small memory allocation of
the test.