GC bug fix for max_collect_interval computation
#45727
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
oscardssmith
approved these changes
Jun 17, 2022
|
Wait. I think you need to make sure that jl_n_threads is properly
initialized. It's done after GC setup by default.
…On Fri, Jun 17, 2022, 15:39 Sacha Verweij ***@***.***> wrote:
***@***.**** approved this pull request.
Thanks Kiran! :)
—
Reply to this email directly, view it on GitHub
<#45727 (review)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ATBN3JQK5DUJWIBO4C3UK4LVPTWCBANCNFSM5ZDMCCEQ>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
Or at least this was the case in 1.7.1
…On Fri, Jun 17, 2022, 16:31 Jan Rous ***@***.***> wrote:
Wait. I think you need to make sure that jl_n_threads is properly
initialized. It's done after GC setup by default.
On Fri, Jun 17, 2022, 15:39 Sacha Verweij ***@***.***>
wrote:
> ***@***.**** approved this pull request.
>
> Thanks Kiran! :)
>
> —
> Reply to this email directly, view it on GitHub
> <#45727 (review)>,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/ATBN3JQK5DUJWIBO4C3UK4LVPTWCBANCNFSM5ZDMCCEQ>
> .
> You are receiving this because you were mentioned.Message ID:
> ***@***.***>
>
|
Not sure exactly when it changed, but GC initialization happens after threads are initialized. |
|
The |
Currently constrained to `totalmem / ncores / 2` for `_P64` which results in a very short collect interval when you're running with a smaller number of threads on a machine with many cores. Changes this to `totalmem / nthreads / 2` which, for two of our tests, resulted in 40% and 60% runtime reduction (!!) as well as GC time reduction from 46% to 10% and 64% to 11%.
|
If this PR causes OOMs, something else is broken. @chflood any thoughts? |
|
32 bit likes to OOM a lot, or more specifically run out of address space so it could be something else. |
|
|
|
Trying to run with this backported to 1.8 I get: I guess cc @kpamnany |
|
I'll check. Should I open a PR to a particular branch? |
|
|
36 tasks
KristofferC
pushed a commit
that referenced
this pull request
Jul 5, 2022
* Bug fix for `max_collect_interval` computation (#45727) Currently constrained to `totalmem / ncores / 2` for `_P64` which results in a very short collect interval when you're running with a smaller number of threads on a machine with many cores. Changes this to `totalmem / nthreads / 2` which, for two of our tests, resulted in 40% and 60% runtime reduction (!!) as well as GC time reduction from 46% to 10% and 64% to 11%. * Move GC init after threading init To allow use of `jl_n_threads` in GC initialization.
pcjentsch
pushed a commit
to pcjentsch/julia
that referenced
this pull request
Aug 18, 2022
Currently constrained to `totalmem / ncores / 2` for `_P64` which results in a very short collect interval when you're running with a smaller number of threads on a machine with many cores. Changes this to `totalmem / nthreads / 2` which, for two of our tests, resulted in 40% and 60% runtime reduction (!!) as well as GC time reduction from 46% to 10% and 64% to 11%.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Currently,
max_collect_intervalis constrained tototalmem / ncores / 2for_P64which results in a very short collect interval when you're running with a smaller number of threads on a machine with many cores.Change this to
totalmem / nthreads / 2which, for two of our tests, resulted in 40% and 60% runtime reduction (!!) as well asGC time reduction from 46% to 10% and 64% to 11%.
Spotted by @janrous-rai.