-
Notifications
You must be signed in to change notification settings - Fork 9
1830: Remove concept of under/overloaded from TemperedWMin
#1860
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
base: develop
Are you sure you want to change the base?
Conversation
Pipelines resultsPR tests (gcc-6, ubuntu, mpich) Build for 6121bc8 PR tests (clang-3.9, ubuntu, mpich) Build for 6121bc8 PR tests (nvidia cuda 10.1, ubuntu, mpich) Build for 3c8faeb PR tests (gcc-10, ubuntu, openmpi, no LB) Build for 6121bc8 PR tests (gcc-9, ubuntu, mpich, zoltan) Build for 6121bc8 PR tests (clang-5.0, ubuntu, mpich) Build for 6121bc8 PR tests (nvidia cuda 11.0, ubuntu, mpich) Build for 6121bc8 PR tests (gcc-7, ubuntu, mpich, trace runtime, LB) Build for 6121bc8 PR tests (clang-9, ubuntu, mpich) Build for 6121bc8 PR tests (clang-13, alpine, mpich) Build for 6121bc8 PR tests (intel icpx, ubuntu, mpich) Build for 6121bc8 PR tests (gcc-8, ubuntu, mpich, address sanitizer) Build for 6121bc8 PR tests (clang-12, ubuntu, mpich) Build for 6121bc8 PR tests (clang-11, ubuntu, mpich) Build for 6121bc8 PR tests (clang-13, ubuntu, mpich) Build for 3c8faeb PR tests (gcc-11, ubuntu, mpich) Build for 6121bc8 PR tests (clang-14, ubuntu, mpich) Build for 6121bc8 PR tests (gcc-12, ubuntu, mpich) Build for cf8804f (2023-02-28 21:41:36 UTC) PR tests (intel icpc, ubuntu, mpich) Build for 6121bc8 PR tests (clang-10, ubuntu, mpich) Build for 6121bc8 PR tests (gcc-5, ubuntu, mpich) Build for 6121bc8 |
Codecov Report
@@ Coverage Diff @@
## develop #1860 +/- ##
===========================================
- Coverage 84.90% 84.37% -0.54%
===========================================
Files 721 760 +39
Lines 25737 26811 +1074
===========================================
+ Hits 21852 22621 +769
- Misses 3885 4190 +305
|
6a85c17 to
d863bd3
Compare
nlslatt
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Within propagateIncomingSync and propagateIncomingAsync there are conditionals on isUnderloaded() that I think should be (effectively) unconditional for TemperedWMin.
You might consider renaming underloaded_ to potential_recipients_ or something like that.
There are configurations possible for TemperedLB that are not valid for TemperedWMin, such as NormByMaxExcludeIneligible for the CMF. We'll need to validate the options at some point but, for now, know that makeSufficientlyUnderloaded should never be used with TemperedWMin.
3c8faeb to
2ae50dc
Compare
f7b4908 to
f363c13
Compare
nlslatt
left a comment
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Your changes to the algorithms to remove the concept of under/overloaded look good. When doing any testing, be sure to use the default CMF NormByMax because the others will not be valid for TemperedWMin.
A good test might be that, with knowledge set to Complete, informSync and informAsync correctly gain knowledge of loads for either all ranks or all underloaded ranks depending if TemperedWMin or TemperedLB is in use. I haven't thought through how to set up such a test yet.
f363c13 to
6121bc8
Compare
6121bc8 to
8ca421f
Compare
742d140 to
5f41170
Compare
5f41170 to
7573b94
Compare
89ea20e to
e2777a3
Compare
e2777a3 to
23608d2
Compare
- extract helper methods and add overloads for TemperedWMin - remove redundant code - remove unused member variable
23608d2 to
cf8804f
Compare
fixes #1830