Bruno.Bossier AT comparex DOT be wrote, in part:
> We have a LAN client which is located at another site several kilometers
> from the site where the master server and the tape library are located.
> There is only a 2Mbit line between the 2 sites, so the backup takes a very
> long time. Actually, until now, it never ran completely because of network
> timeouts or because it was out of the time window or because it was
> cancelled by the owners.
>
> What can I do to improve at least a little the performance ? I already
> thought of disabling OTM, setting the communications buffer size to 256
> (now 32). ANother thought would be to enable compression at the policy
> level. I know what the advantages and disavantages are of using software
> compression. What I would like to know is what happens if you use software
> compression and hardware compression at the same time ?
I'm new to NBU, but I'll offer ...
1. Don't send as much. Perhaps there are large static files that can be
backed up less frequently? Consider daily backup for only most
important files and every other day (or week) for rest, breaking up
remaining between the two days. If backups are frequently failing with
time going unused, consider multiple policies so that you've got a bunch
of small restartable backups instead of one or a few large backups.
2. Send it faster. Sometimes us software folks get caught up with
software issues when it's a hardware problem. First, have a networking
guru study the line (under load); there may be adjustments s/he can make
to get more out of your 2Mb connection. Second, consider the need for a
faster connection and possible solutions. The guru just mentioned may
be able to provide all the info you need, or at least get you started.
While it may be worth having NBU compress your data, I've yet to find a
real life case where it helps. If you have enough (remote) computing
power and NBU is good enough at overlapping disk reading, compressing
and transmission (it isn't, in my short experience), it might help.
That your tape "hardware" also does compression just doesn't matter to
this problem at all; it's not a problem in any sense.
If you change NBU settings (NET_BUFFER_SZ, NUMBER_DATA_BUFFERS,
SIZE_DATA_BUFFERS, NUMBER_DATA_BUFFERS_RESTORE, TCP/IP stack) and see a
difference (or don't!), please report here. IMHO, Veritas and we need
to discuss this area (and make improvements) more (Yes, I've read
everything in sight!). Some areas that I find incredible: (1) Veritas
suggests "testing" since changing some of these values may cause
restores to fail. (2) Veritas suggests some settings must be made at
server and all clients; if a setting requires this, NBU should reflect
the setting to the clients and use it automatically. (3) transmission
buffers, NBU buffers, and tape I/O buffers aren't distinct and
independent. (4) NBU apparently is unable to tune itself, simply using
"one size fits all".
Has anyone written scripts to analyze NBU logs wrt performance issues?
Are there any add-on products for this (or other) area(s)?
(Must stop now ... have worn out parenthesis keys).
cheers, wayne
--
Wayne T. Smith -- WTS AT Maine DOT edu
University of Maine System -- UNET -- Systems Software Analyst
|