Re: journaling filesystem

ralf@uni-koblenz.de
Mon, 18 May 1998 10:29:54 +0200


On Mon, May 18, 1998 at 03:56:45AM -0400, Chris Siebenmann wrote:

> You write:
> | I didn't do any quantitive measurements on this but my feeling says
> | that the two ``undestructibe'' filesystems I've worked with (Solaris +
> | Solstice Disksuite and IRIX's XFS) both have reduced write performance
> | with the log on the data disk.
>
> In not particularly scientific tests, XFS appears to turn in streaming
> write speeds (tested via bonnie) of about the same or slightly faster
> than SGI's EFS on otherwise identical hardware and software setups.

This is probably true since XFS's code is said to be better implemented
and definately has a better block allocation policy. Both EFS and XFS
are extend based, so the amount of metadata logging is minimal for
streaming writes anyway. On simple configurations usully the hardware is
the limit for every not completly stupid fs implementation and so it's
pretty impossible to be much faster than EFS. Things are different for
_big_ configurations where XFS can deliver 3gb/s r/w sustained.

Somebody promised to finally fix the EFS code for Linux, so we should
be able to see how EFS does compared to ext2 in the same environment.

It's still funny to see how the dcache is able to blow IRIX away for
certain important real live applications like "make distclean". On the
same hardware Linux + ext2 with cold buffercache is over 4 times as fast
as IRIX 6.2 + XFS with hot buffercache. FS performance isn't everything,
the battle is won as close to userspace as possible and the dcache does
a pretty good job at it.

Ralf

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu