[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Reply: bugzilla-noreply_a_freebsd.org: "[Bug 270785] Performance and power efficiency regression due to pthread_cond_timedwait() changes"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Wed, 12 Apr 2023 13:08:57 UTC
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=270785
Bug ID: 270785
Summary: Performance and power efficiency regression due to
pthread_cond_timedwait() changes
Product: Base System
Version: 13.2-RELEASE
Hardware: Any
OS: Any
Status: New
Severity: Affects Some People
Priority: ---
Component: kern
Assignee: bugs@FreeBSD.org
Reporter: bllcha013@myuct.ac.za
After updating from 13.1-RELEASE-p7 to 13.2-RELEASE-p0, I experienced a
performance and power efficiency regression especially when running Java
programs, in particular the Minecraft server.
After some testing, this appears to be caused by commit
e21c77f80c3b1cf2bd0a9f874e15e7a8e49f0dba [Improve timeout precision of
pthread_cond_timedwait().]. When building the kernel just before this commit,
this regression does not occur, while building with this commit
(n253051-e21c77f80c3b), the regression occurs. The regression is still present
in the 6 April 2023 snapshot kernel of 14.0-CURRENT.
I have observed this behavior on two systems, a server with an Intel Xeon X3460
(Lynnfield, Nehalem) CPU, as well as a desktop PC with an Intel Core i7-6700K
(Skylake) CPU.
When running a kernel with this commit, the WCPU usage of the java process in
top is considerably higher at idle than without the commit when running a
Minecraft server (1 to 2% before, 10% to 30% after). In addition, the CPU
spends a lot more time in less power efficient C-states (C1) where before this
commit, the CPU spends much more time in power efficient C-states such as C3.
On my server, the CPU used to spend 98% time in C3 while after the commit it
spends only 30% time in C3, leading to higher power consumption, more heat, and
faster fan speeds.
--
You are receiving this mail because:
You are the assignee for the bug.