All,
I have been using the new LVM snapshot regression test extensively over the
last couple of business days. (xfstests/068 now in CVS)
I have found that the only kernel that can reliable pass the test is an xfs cvs
kernel as of today. (Steve Lord put in a fix over the weekend that seems to be
the final piece in the puzzle.)
I suggest anyone using LVM snapshots with XFS in a production environment
should consider this information.
Note: if you want to run the snapshot test on your platform, the current CVS
version has a couple of small typos that I described in an e-mail earlier
today. Hopefully Nathan will get those changes into CVS shortly.
Note 2: The kernel lockup listed below has not been reported before by me, but
it was apparently resolved by Steve's latest fix.
===== Results of my tests
WIth no i/o load:
Kernel 2.4.19pre4aa1with SuSE patches (i.e. XFS 1.1 based) Succesfully
ran 20 iterations
XFS CVS Aug. 30 kernel
Succesfully ran 20 iterations
XFS CVS Sept. 2 kernel
Succesfully ran 20 iterations
WIth a dd based i/o load
Kernel 2.4.19pre4aa1with SuSE patches (i.e. XFS 1.1 based) lvcreate
locks up within 20 iterations
XFS CVS Aug. 30 kernel
Succesfully ran 20 iterations
XFS CVS Sept. 2 kernel
Succesfully ran 20 iterations
WIth a fsstress based i/o load
Kernel 2.4.19pre4aa1with SuSE patches (i.e. XFS 1.1 based) lvcreate
locks up within 20 iterations
XFS CVS Aug. 30 kernel
Not tested
XFS CVS Sept. 2 kernel
Succesfully ran 20 iterations
WIth both a dd and a fsstress based i/o load
Kernel 2.4.19pre4aa1with SuSE patches (i.e. XFS 1.1 based) lvcreate
locks up within 20 iterations
XFS CVS Aug. 30 kernel
Kernel locks up within 20 iterations
XFS CVS Sept. 2 kernel
Succesfully ran 20 iterations
Greg Freemyer
Internet Engineer
Deployment and Integration Specialist
Compaq ASE - Tru64 v4, v5
Compaq Master ASE - SAN Architect
The Norcross Group
www.NorcrossGroup.com
|