Steve Giessler
<sgiessle AT wvu DOT edu> To:
veritas-bu AT mailman.eng.auburn DOT edu
Sent by: cc:
veritas-bu-admin AT mailman DOT eng. Subject:
[Veritas-bu] multiple streams/ALL_LOCAL_DRIVES/slow restores
auburn.edu
05/31/2002 02:51 PM
Hi,
I'm new to Netbackup and need a few pointers if someone would be so kind.
I have NB version 3.4.1 running on a Windows 2000 box. I have a
Cybernetics 15 tape robotic library connected to that box with a
single SONY SDX-500C drive in it (uses 50 gig native AIT2 tapes).
I have three Solaris 8 clients that I want to backup. I want them all
to multiplex to one tape in as many streams as possible at one time
(8 seems to be the limit). My understanding is that if I choose
ALL_LOCAL_DRIVES, and have multiplexing and allow multiple streams
selected (which I do and both are set to 8), then each mount point
will initiate a separate network stream of data to the drive and up
to 8 streams will backup concurrently.
Is this correct?
If so, it is not working this way. What I get when I start a backup
is this: The mount points are each "Queued" and only one stream is
started from each host. When the single stream on a given host is
finished, it begins the next queued mount point on that host.
Shouldn't it be able to stream multiple mount points at the same time
from a single host?
>> you'll have to either change the max jobs per client global attribute or
configure the max jobs per client higher for this particlular client
In any case - each of my 3 Solaris client machines contains at least
3 hard drives in them - each with separately mounted partitions and
mount points on each drive. So, at a minimum, I see a potential for 9
streams right there. Why then, when I run a backup, can I only get 2
streams at a time to run and not the desired 8? . This way it will
take forever to backup all that data.
>> in addition to max jobs per client, you'll want to play with the max
mpx/schedule and max mpx/storage unit (to keep the drives streaming) and
number of concurrent drives per storage unit (to keep from flooding one media
server while other available media servers are idle)... also keep in mind that
by default retention levels won't mix on same media.
Another question I have is in regard to the ALL_LOCAL_DRIVES directive:
It seems that when I do a backup with the ALL_LOCAL_DRIVES directive
selected, I am getting redundant backups. What I mean is when I do a
full backup, each mount point gets backed up, and then when it backs
up / (root) partition - it goes through and backs up all of the
subdirectories under root also - in other words - all of my mount
points - the entire file system gets backed up - twice in one full
backup.
How can I prevent this? I don't want to have to manually tell it
which directories to backup (never had to with Legato Networker).
>> I think if you use / in the file list, instead of ALL_LOCAL_DRIVES, and
allow it to cross mount points and do multiple data streams, you'll get the
results you're looking for. There have been a lot of other posts talking
about the difference between the two, and which people prefer, if you want to
search through the archives for it.
I have used Legato Networker on Unix in the past and so far, I wish I
hadn't switched, but hope that it is just my ignorance and needing to
work out the details.
My last question is this: When I do a restore - it seems to read
through the entire 50 gig tape in order to find the files that were
selected to restore. With Legato Networker, the index files kept
track of the position of the files on the tape and it would find and
restore them very quickly (in 5 minutes or less usually). I tried
restoring a few small files (200 K or so total) and it took about an
hour for it to scan through the entire tape before it could find the
files and restore them. Surely I must be doing something wrong. I
have two catalog locations selected - one is the first tape in my
library, and alternately a directory on my master server's hard disk.
>> set the fragment size for the storage unit to 1024 or 2048 instead of the
default 0. This will write multiple smaller tarballs instead of one large
one. The restores will position to the correct tarball and read through less
to get to the files.
>> the two locations are your catalog backups and they will aternate for each
catalog backup taken. You don't want one on your server's hard disk unless
it's in another location across a SAN. If you lose this server along with
this backup on disk, and it is the most recent one used, the copy on the tape
will not have everything you want. The image catalog is in
netbackup/db/images/* on the master.
Any suggestions greatly appreciated,
Thanks,
--
Steve Giessler, Unix System Administrator
West Virginia University
Eberly College of Arts and Sciences
201 Woodburn Hall
Morgantown, WV 26506
(304) 293-7855 x2214 or
(304) 293-2011 x2214
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
|
