Date: Sun, 23 Sep 2001 14:22:11 -0700 From: Julian Elischer <julian@elischer.org> To: Matt Dillon <dillon@earth.backplane.com> Cc: Poul-Henning Kamp <phk@critter.freebsd.dk>, David Greenman <dg@root.com>, Seigo Tanimura <tanimura@r.dl.itc.u-tokyo.ac.jp>, bright@wintelcom.net, hackers@FreeBSD.ORG Subject: Re: Conclusions on... was Re: More on the cache_purgeleafdirs() routine Message-ID: <3BAE5283.C7B434CF@elischer.org> References: <1620.1001272770@critter> <200109232019.f8NKJZL88354@earth.backplane.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Matt Dillon wrote:
>
> :Block input operations is the one notable exception and it tells a
> :very interesting story: Matts patch results in a 4% increase, but
> :combined with vmdirioenable it results in a 21.5% decrease.
> :
> :That's pretty darn significant: one out of every five I/O have
> :been saved.
Notice that both the user and system times increased..
if there had been another parallel task, the overall system throughout may have
decreased..
I'm not saying this is wrong, just that we should look at other workloads too.
no point in optimising the system for compiling itself.. that's not really a
real-world task..
lmost noo-one buys a freeBSD box so that they can compile freeBS.. they usually
have other plans for what they want out of it...
(of course this may effect other tasks in an even more positive way, but we
should know that..)
> :
> :The reason it has not manifested itself in the "real" number is
> :probably the high degree of parallelism in the task which practically
> :ensures that the CPU will not go idle.
> :
> :I suggest we let Matt's patch depend on the vmiodirenable sysctl
> :and change the default for that.
> :
> :If there are no bad side effects found in the next couple of months,
> :then kill the sysctl and lets be done with it.
> :
> :Poul-Henning
>
> Very interesting! I agree completely. I will instrument the code
> to make the namei cache check conditional on vmiodirenable. I will
> also add a sysctl to change the check condition to test for leaf
> directories or not (closer to what Seigo's code accomplished) so we can
> test for differences there. If we can get that 4% down the code will
> be more acceptable.
>
> The results make a lot of sense, especially if the machine has sufficient
> memory to cache most of the source tree. Increasing the number of
> vnode reclaims will increase I/O on directories, especially for something
> like a buildworld which is doing major path expansions up the wazoo.
> Turning on vmiodirenable in a buildworld situation will decrease I/O
> significantly.
>
> -Matt
>
> To Unsubscribe: send mail to majordomo@FreeBSD.org
> with "unsubscribe freebsd-hackers" in the body of the message
--
+------------------------------------+ ______ _ __
| __--_|\ Julian Elischer | \ U \/ / hard at work in
| / \ julian@elischer.org +------>x USA \ a very strange
| ( OZ ) \___ ___ | country !
+- X_.---._/ presently in San Francisco \_/ \\
v
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?3BAE5283.C7B434CF>