I have come across issues with our disk staging units as
well, but it is not fragmentation...
out of two media servers, only one presents the symptom of
de-staging slowing down.
System A - Master Server, Solaris 10 Sparc, 2x DSU's (each
4x 4+1 FC attached SATA disks) formatted VXVM/VXFS.
System B - Media Server to System A, Solaris 10 Sparc, 1x
DSU (4x 4+1 FC attached SATA disks) formatted ZFS.
System B gives us consistent performance, month in month
out.
System A works great, better than System A (by about
50MB/s). However, there is a noticeable decrease in de-staging performance
(probably staging performance as well, but who is up at 11pm to see
that) after about 2 weeks of system uptime. and after about 4 weeks the
issue is blatantly staring us in the face, we see about only 50MB/s of
de-staging performance.
The last time this issue hit us I did some tests, dd from
DSU to tape = Slow, dd from DSU to null = Slow, dd from zero to DSU = Slow, DD
from Zero to tape = FAST :-). ok, I'm happy it is not tape subsystem, I thought
for a moment I have a ZFS mirror of 2x 2.5" SAS disks on System A, copied a
Large image from DSU to the SAS disk. I performed first the dd test of that
image to null from the DSU, still slow. OK, I performed the dd test from SAS to
null = Fast. Wait, What a single 2.5 SAS disk is faster that 16x SATA disks ? No
way.
I knew through previous experiences that a reboot clears
(temporarily) the symptom. and on this most recent experience I rebooted,
and as soon as it returned, dd from the DSU to null = Blazingly
Fast.
I will be replacing the VXFS DSU's with ZFS in the
future.
Regards
Adam.
It's not just a fragmentation issue - a defrag'ed file system may
also give you poor destage performance.
A bunch of people have reported
this issue but so far nobody has found the smoking gun. It's an issue on
multiple platforms as well.
.../Ed
On Wed, May 6, 2009 at 4:10 AM, owiking <netbackup-forum AT backupcentral DOT com>
wrote:
Im
having problems with my diskstage getting to fragmented and thus getting very
slow performance when duplicating to tape. Running defrag isnt an option since
the files are reaplced fequently and the defrag job will take to long time to
run. From what Ive seen and read, this is a major problem for alot os users
of Netbackup.
Is there a way to force Netbackup to preallocate space on
the diskstage for each backupjob, instead as it is now that they are all
written in fragments all over the disk?
Ive read some about the
differences with Basic and Advanced diskstages in Netbackup, Im running Basic
atm, but havent found any information that Advanced could preallocate
space?
My diskstage is 5.5tb large, is it better to make to divide it
to 2-3 partitions to avoid fragmentation?
.../Ed
Ed
Wilts, RHCE, BCFP, BCSD, SCSP, SCSE ewilts AT ewilts DOT org
NOTICE: This email and any attachments are confidential.
They may contain legally privileged information or
copyright material. You must not read, copy, use or
disclose them without authorisation. If you are not an
intended recipient, please contact us at once by return
email and then delete both messages and all attachments.
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
|