Yes, they dont solve this whatsoever.  Its a severely broken code
issue in the attr caching mechanism.

Honest..give it a shot.  Even with very very slow ATA, local builds of
the kernel or world are faster than over 1G NFS to an F6000 series
filer, and the filer will still thrash on WAFL metadata requests for
the client, cuz everytime ../.. walks somewhere the client knows
nothing and it's all sent over the wire again.  Over and over and
over.

Thats as much as I understand about it, freebsd-fs has great detail on
this "bug".

On 1/11/07, Freminlins <[EMAIL PROTECTED]> wrote:
On 11/01/07, Jeff Mohler <[EMAIL PROTECTED]> wrote:
>
> The basic reason is that a ../.. walk invalidates cached metadata, and
> you end up with a pipe full of getattr's all of the time.  Freebsd-fs
> has discussed this a bit, but no fixing is coming soon.  We use linux
> to compile builds, we'd like to use Freebsd, but linux on Filers via
> NFS is about 3x faster than the same builds on Fbsd to the same filer.
>   ../.. baby.
>


Did you try different mount options on the FreeBSD clients. I have no idea,
but Linux may have different defaults.

Frem.

_______________________________________________
freebsd-questions@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-questions
To unsubscribe, send any mail to "[EMAIL PROTECTED]"

Reply via email to