Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 20 Jan 2002 13:02:44 -0600
From:      Dan Nelson <dnelson@allantgroup.com>
To:        Poul-Henning Kamp <phk@critter.freebsd.dk>
Cc:        Duraid Madina <duraid@fl.net.au>, freebsd-hackers@FreeBSD.ORG
Subject:   Re: Insane performance regression?
Message-ID:  <20020120190244.GF81627@dan.emsphone.com>
In-Reply-To: <2427.1011528090@critter.freebsd.dk>
References:  <000001c1a16a$ec95cc50$022a17ac@simplex> <2427.1011528090@critter.freebsd.dk>

Next in thread | Previous in thread | Raw E-Mail | Index | Archive | Help
Bad top-poster, no cookie.

In the last episode (Jan 20), Poul-Henning Kamp said:
> In message <000001c1a16a$ec95cc50$022a17ac@simplex>, "Duraid Madina" writes:
> >	I have a CPU-bound (well, 'malloc-bound' ;) program which takes
> > about 20 seconds to run on a 'fast' PC (Pentium3-1000, Athlon
> > XP1600 etc) - the source is available as
> > http://www.idesign.fl.net.au/malloc_pain/malloc_pain.tar.gz (NOTE:
> > you *will* need GCC 3 (or more recent) to compile it). At any rate,
> > I did the cvsup/buildkernel/buildworld thing this morning (I'm
> > running 5-CURRENT on an SMP box), and now that same program takes
> > about half an hour to run, rather than 20 seconds. Curiously, it
> > reports about 20% system time (whereas previously there was
> > negligible system time)
>
> One, check your malloc option settings.  FreeBSD-current defaults to the
> AJ setting to flush out errors but this has a significant performance
> hit.

Duraid: were you running 4.* before and just upgraded to -current, or
did you simply bring an older -current box up to date?  If the latter,
you might want to try building different kernels to see if you can
pinpoint the commit that's causing your slowdown.  You will only have
to rebuild the kernel, and a binary search should let you narrow it
down in under 2 hours, I'd say.  I would have suggested looking at
Alfred's SMP file locking commit on 2001-01-13, but if your program
just does malloc()s it shouldn't be affected by that.

-- 
	Dan Nelson
	dnelson@allantgroup.com

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message




Want to link to this message? Use this URL: <http://docs.FreeBSD.org/cgi/mid.cgi?20020120190244.GF81627>