Date: Fri, 1 Jun 2001 19:03:48 -0700 (PDT) From: Matt Dillon <dillon@earth.backplane.com> To: Terry Lambert <tlambert2@mindspring.com> Cc: Robert Watson <rwatson@FreeBSD.ORG>, Ian Dowse <iedowse@maths.tcd.ie>, freebsd-hackers@FreeBSD.ORG Subject: Re: UFS large directory performance Message-ID: <200106020203.f5223mZ97356@earth.backplane.com> References: <Pine.NEB.3.96L.1010601123814.65702E-100000@fledge.watson.org> <200106011806.f51I6PK85431@earth.backplane.com> <3B18338C.4641B4F4@mindspring.com>
next in thread | previous in thread | raw e-mail | index | archive | help
:The new code is interesting; it will be enlightening to
:see it's real world performance. I'd definitely suggest
:using a zone for the allocations, however.
:
:FWIW: I guess if you are having problems with mail queue
:perofrmance, you are running postfix or qmail or something,
:instead of sendmail, with the mail queue divisions, or with
:my and David Wolfskill's per-domain mail queue patches?
:
:-- Terry
No problems at Backplane. I was speaking historically. The
multi-queue stuff certainly helps, I hacked up a multi-queue
sendmail setup at BEST Internet, but it still wasn't perfect. It just
changed O(X^2) to O(Y * [X/Y]^2) (e.g. try with X=1000 and Y=10).
A definitely improvement, but Ian's stuff can get it down to O(Y) for
all intents and purposes.
There is still sheer drop in regards to scaling Ian's solution after
a directory grows past a few hundred thousand files, but I'm not too
worried about it. The memory use ratio is good enough that that adding a
little memory to a machine lets you pile on a whole bunch more files
in the hash solution.
-Matt
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: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200106020203.f5223mZ97356>