List:General Discussion« Previous MessageNext Message »
From:Dan Nelson Date:January 21 2003 3:43pm
Subject:Re: FreeBSD 5.0 Thread Work...
View as plain text  
In the last episode (Jan 21), Ken Menzel said:
>    Thanks for that information,  does that mean (for us simple SQL
> guys) that even though I am running 5.0RC3 and can compile and
> install MySQL just fine (from ports or with my own script) it is
> using the threads library from 4-STABLE branch not the new threads?

It is still using userland threads, yes.  You're limited to one CPU and
threads will have to wait for each other's disk I/O.
 
> Also does this have soemthing to do with the word GIANT0 that keeps
> appearing in the 'top' process list?

Where in 4.* all system calls implicitly grabbed a kernel lock to keep
multiple processes from entering the kernel at the same time, about
half of 5.*'s syscalls are multi-processor safe.  The unsafe ones still
have to grab a lock, but in 5.* it's called Giant and it shows up in
top.
 
> <observations>
> I was hoping for great things from the 5.0, even with kernel profiling
> off (invariants?) the /usr/local/sql-bench runs slower on 5.0.  It is

WITNESS is the kernel option that's most likely slowing you down here. 
It forces each lock attempt to prove that it won't cause a deadlock.
INVARIANTS doesn't hurt too much.  Also try disabling the J malloc
debug flag by running ln -s 'j' /etc/malloc.conf.

-- 
	Dan Nelson
	dnelson@stripped
Thread
FreeBSD 5.0 Thread Work...Peter Grigor20 Jan
  • Re: FreeBSD 5.0 Thread Work...Maximo Migliari20 Jan
    • Re: FreeBSD 5.0 Thread Work...Dan Nelson20 Jan
  • Re: FreeBSD 5.0 Thread Work...Ken Menzel21 Jan
    • Re: FreeBSD 5.0 Thread Work...Dan Nelson21 Jan
  • Re: FreeBSD 5.0 Thread Work...Jeremy Zawodny21 Jan
  • Re: FreeBSD 5.0 Thread Work...Ken Menzel23 Jan
    • Re: FreeBSD 5.0 Thread Work...Jeremy Zawodny24 Jan
RE: FreeBSD 5.0 Thread Work...Peter Grigor21 Jan