Comments on: Fix a bug in Ruby’s configure.in and get a ~30% performance boost.

By: Jeff Blaine

Jeff Blaine — Sat, 05 Mar 2011 02:23:51 +0000

This is not fixed in 1.8.7 p330 (latest 187 2 months ago). Comment 3 of mine: http://www.ruby-forum.com/topi...

"""
Solaris 10 on SPARC, Ruby 1.8.7 p330, --enable-pthread

Ruby application 'puppet'

"puppet resource user root" issues over 2 million getcontext() calls and
takes 25 seconds to complete

On a Linux box, same build args, the command takes 1-2 seconds.
"""

By: teh lix

teh lix — Wed, 09 Feb 2011 01:01:05 +0000

Same story on FreeBSD:

3049 ruby18 CALL _umtx_op(0x8017d1060,0x8,0x1,0x8017d1040,0x7fffe7bfdec0)
3049 ruby18 RET _umtx_op -1 errno 60 Operation timed out
3049 ruby18 CALL clock_gettime(0,0x7fffe7bfdf80)
3049 ruby18 RET clock_gettime 0
3049 ruby18 CALL clock_gettime(0,0x7fffe7bfded0)
3049 ruby18 RET clock_gettime 0
3049 ruby18 CALL _umtx_op(0x8017d1060,0x8,0x1,0x8017d1040,0x7fffe7bfdec0)
3049 ruby18 RET _umtx_op -1 errno 60 Operation timed out
3049 ruby18 CALL clock_gettime(0,0x7fffe7bfdf80)
3049 ruby18 RET clock_gettime 0
3049 ruby18 CALL clock_gettime(0,0x7fffe7bfded0)
3049 ruby18 RET clock_gettime 0
3049 ruby18 CALL _umtx_^Cop(0x8017d1060,0x8,0x1,0x8017d1040,0x7fffe7bfdec0)

By: Unix signal programming in Ruby « Spakman loves triangles

Unix signal programming in Ruby « Spakman loves triangles — Tue, 27 Apr 2010 11:16:38 +0000

[...] Joe Damato found, MRI 1.8 with pthreads enabled is rather rt_sigprocmask() happy. It seemed obvious that all that [...]

By: lazzarello

lazzarello — Sat, 24 Apr 2010 01:58:35 +0000

Heya. I applied your patch to ruby 1.8.7 (2008-08-11 patchlevel 72) [x86_64-linux] in Debian stable and rebuilt the package. It does not remove the dominance of the rt_sigprocmask syscall in your test script. I ran the script for 24 seconds and it called that syscall 416614 times.

By: cryptoromantic

cryptoromantic — Thu, 25 Feb 2010 10:04:16 +0000

so what is the underlying bug here – that get/setcontext is completely unnecessary because green threads are being used even with pthreads enabled, or that some work get/setcontext is doing is necessary, but it also always calls sigprocmasks (which is unnecessary), and this calls for a better decomposition of the get/setcontext call… ?

By: Seth Thomas Rasmussen

Seth Thomas Rasmussen — Tue, 02 Feb 2010 01:22:34 +0000

Great work! I've been investigating these issues around –enable-pthread recently as I *just* became aware of their significance.. and I'm surprised, frankly.. there seem to be a lot of people that knew this, yet a lot of people who still don't and who still run into tutorials like the popular Hivelogic one which recommends you use –enable-pthread.

By: staffing1

staffing1 — Thu, 10 Dec 2009 00:30:41 +0000

Brilliant work! I hope the guys upstream patch it in.
www.staffingpower.com

By: staffing1

staffing1 — Wed, 09 Dec 2009 16:30:41 +0000

Brilliant work! I hope the guys upstream patch it in.
www.staffingpower.com

By: nigelkersten

nigelkersten — Thu, 05 Nov 2009 15:41:57 +0000

According to Matz, this isn't a problem for Ruby 1.8.7

http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ru...

yet I'm still seeing performance improvements on the ruby benchmark suite when applying your patch…

By: redditrebel

redditrebel — Thu, 08 Oct 2009 20:59:44 +0000

XBMC and Boxee(based on XBMC) suffer from this on newer 64 bit kernels, calling rt_sigprocmask a bazillion times per second and basically eating a whole CPU.

I stopped trying to debug it as the developers were truculent about even supporting newer 64 bit installs, ever.

I wouldn't be surprised if this is the cause.