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

Decrease the number of send_pending_parcels threads #339

Closed
brycelelbach opened this Issue Jul 10, 2012 · 4 comments

Comments

Projects
None yet
1 participant
@brycelelbach
Member

brycelelbach commented Jul 10, 2012

[reported by blelbach] [Trac time Tue Feb 28 14:27:08 2012] |||GCC 4.6.2, 4b99001, Boost 1.49.0, debug

There are way too many send_pending_parcels threads. Running sheneos_test on two localities with default parameters creates about 27k pxthreads on the console, 11k of which are send_pending_parcels threads.

Steps to reproduce: Run sheneos_test on two localities with default parameters. Use logs to count number of threads like so:

cat my_hpx_log | grep thread::thread | wc -l
cat my_hpx_log | grep thread::thread | grep send_pending_parcels | wc- l
@brycelelbach

This comment has been minimized.

Show comment
Hide comment
@brycelelbach

brycelelbach Jul 10, 2012

Member

[comment by heller] [Trac time Tue Feb 28 14:33:39 2012] In order to determine why, and if, send_pending_parcels are bad, we need network performance counters to fine tune the current connection caching, and parcel buffering.

Member

brycelelbach commented Jul 10, 2012

[comment by heller] [Trac time Tue Feb 28 14:33:39 2012] In order to determine why, and if, send_pending_parcels are bad, we need network performance counters to fine tune the current connection caching, and parcel buffering.

@brycelelbach

This comment has been minimized.

Show comment
Hide comment
@brycelelbach

brycelelbach Jul 10, 2012

Member

[comment by blelbach] [Trac time Tue Feb 28 15:19:36 2012] I don't see how this has anything to do with network performance counters, just decrease then number of send_pending_parcels threads.

In release, it's just as bad. 21k pxthreads, 9k of which are send_pending_parcels.

Member

brycelelbach commented Jul 10, 2012

[comment by blelbach] [Trac time Tue Feb 28 15:19:36 2012] I don't see how this has anything to do with network performance counters, just decrease then number of send_pending_parcels threads.

In release, it's just as bad. 21k pxthreads, 9k of which are send_pending_parcels.

@brycelelbach

This comment has been minimized.

Show comment
Hide comment
@brycelelbach

brycelelbach Jul 10, 2012

Member

[comment by blelbach] [Trac time Tue Feb 28 15:25:10 2012] BTW, in both debug and release I only send about 8k parcels total.

Member

brycelelbach commented Jul 10, 2012

[comment by blelbach] [Trac time Tue Feb 28 15:25:10 2012] BTW, in both debug and release I only send about 8k parcels total.

@brycelelbach

This comment has been minimized.

Show comment
Hide comment
@brycelelbach

brycelelbach Jul 10, 2012

Member

[comment by heller] [Trac time Thu Mar 8 16:59:52 2012] This was fixed with e3f1307.

Member

brycelelbach commented Jul 10, 2012

[comment by heller] [Trac time Thu Mar 8 16:59:52 2012] This was fixed with e3f1307.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment