Re: NFS client performance 1.5 orders of magnitude too slow? (fwd)

Alan Cox (alan@lxorguk.ukuu.org.uk)
Tue, 9 Mar 1999 21:45:57 +0000 (GMT)

> Then I guess whenever this gathering code makes into some newer version of
> linux, I'll have to add another piece of code to amd that says, "if kernel
> version is greater than 2.2.N, then set rsize=wsize=8k". For now I can
> leave it up to people to set these options by hand in their amd maps.

You only want 8K if you are talking to a machine with the BSD vfs [1]
and 8K page size - otherwise 4K is faster. At that point you are really into
"user tune it right" not into making amd guess it right. I don't think picking
4K and forcing the user to set 8K when talking to Solaris sparc, and the like
is a bug.

[1] BSD systems are only generally capable of doing page sized I/O and NFS I/O
is synchronous. The page sized I/O is not an issue except when dealing
with an awkward NFS client. BSD libc's and apps naturally use page or
multipage I/O and its a win for lots of other stuff. When it combines with
the BSD UFS fragment behaviour you really don't want to be doing synchronous
I/O under page size.

Alan

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/