Date: Mon, 26 Jul 1999 08:51:05 -0700 (PDT) From: Matthew Dillon <dillon@apollo.backplane.com> To: "Brian F. Feldman" <green@FreeBSD.ORG> Cc: Alan Cox <alc@FreeBSD.ORG>, cvs-committers@FreeBSD.ORG, cvs-all@FreeBSD.ORG Subject: Re: cvs commit: src/sys/kern vfs_bio.c vfs_lookup.c vfs_subr.c vfs_syscalls.c vfs_vnops.c src/sys/sys vnode.h Message-ID: <199907261551.IAA46895@apollo.backplane.com> References: <Pine.BSF.4.10.9907260326500.15081-100000@janus.syracuse.net>
next in thread | previous in thread | raw e-mail | index | archive | help
:Does this mean that you want us to try it out and get data points for
:the problems Matt found with VMIOed directories?
:
: Brian Fundakowski Feldman _ __ ___ ____ ___ ___ ___
: green@FreeBSD.org _ __ ___ | _ ) __| \
Yes. Anyone who wants to play with VMIO'd directories can now.
The vfs.vmiodirenable sysctl defaults to 0 (off).
We need to work up two tests:
* First, the effect VMIOing directories has on the VM cache. We know for
a fact that in a directory-intensive environment directories will
stick around in the VM cache much longer then they previously stuck
around as B_MALLOC buffers in the buffer cache. The real question is
whether the additional 'waste in space' creates a problem or not.
Smaller (< 1K) directories ate 1K as B_MALLOC buffers but will eat 4K
as VMIO-backed buffers.
* Second, the last time we tried to turn VMIO'd directories on
softupdates started panicing occassionally. We need to determine
if this still occurs and, if it does, try to collect information that
will enable Kirk to track the problem down after he gets back from
his vacation.
The best environment for both tests is a relatively busy server running
on a large dataset. If the server pages normallly the effect of #1 can
be seen most easily by determining whether the server pages more or less
due to the enabling of VMIO'd directories.
-Matt
Matthew Dillon
<dillon@backplane.com>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199907261551.IAA46895>