I know this isn't a super-scientific test, but, I wanted to get an
idea of what sort of write performance changes you see on Solaris
with different filesystems, using, or not using fsync()'ing all over
the fileserver...
Test System: Dell 2650, 8G ram, local SCSI disk
OS: Solaris 10u2 x86 (32bit mode...ugh)
AFS: OpenAFS 1.4.1 (both client & server), namei fileserver
Cache: 2G RAM cache (lofi'd ufs filesystem cache)
Test: Very simple. Untar gcc-4.1.1 into a virgin volume, on a virgin
fs, rebooting
between test to clear the memory cache...
backing store fsync'ing no-fsync'ing
---------------------------------------------------------
logging ufs 8min54sec 2min36sec
zfs n/a 1min46sec
zfs w/compression n/a 1min59sec
I didn't bother testing zfs with the fsync() on, I knew it would be
pretty horrible in light of the current fsync "bug" it's got...
Yes, that really is a 4x performance boost on writes. w00t, as the
kids say...
-rob
_______________________________________________
OpenAFS-devel mailing list
[email protected]
https://lists.openafs.org/mailman/listinfo/openafs-devel
