We did some inters testing testing yesterday.
The SAN admin has zoned all servers to see all tape
drives. So now I'm looking at the possibility of one or more servers being
the cause, with either polling or SCSI reserve. There's a well-known issue
with HP's EMS polling which I've double-checked, but the SCSI reserve might be
it.
We have one solo tape drive that is only zoned in the SAN
to two systems and we did a speed test yesterday. The exact same
backup to the non-shared drive was about 8x faster than to a tape drive zoned
across all media servers
When configuring drives does anyone
choose an option other than default system SCSI reserve? I would
presume that if this is done (as we have it) then the guidelines in the Device
Config pdf need to be adhered to (we are on HP, not sure about Solaris,
etc.) Is it better to disable SCSI reserve on all the hosts and use
Netbackup's SPC2 methodology? Or to take the steps outline to enabled
persistent SCSI reserve as outlined in the Device Config
guide?
So since we're using default system scsi reserve setting for each drive,
it looks likely a system, or set of systems, is causing issues by interrupting
tape drives.
We have all
our tape drives (20) assigned to all our media servers (3 main ones, and ~45
that server as SAN media servers, backing only themselves
up).
Dealing with
tape drive throughput issues and we've had a Brocade support person tell us that
this setup is not ideal and that even when a tape drive has been granted to a
host and it starts to use it, that its checking on the other tape drives
prevents good throughput. He says to only assign about 4 drives per media
server.
This does
not seem right to me, but hey, I'll check out anything. My understanding
is that 6.5 (we're at 6.5.2a, about to go to 6.5.3) has a centralized device
database and that the design of a SCAN_HOST performs the gatekeeper roll for the
tape drives.
I've added a
new server and only configured 4 drives to run a test on it. Of course it
is queued up waiting for a free tape drive, and there's been 147 other jobs that
got a tape drive while it's been queued up...(proving point #1 for why all
servers had seen all drives).
If you have
any opinions on Brocade's theory (or at least our SAN guy's interpretation of it
- this is all 2nd hand) I'd be happy to receive
it.
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
|