Data Storage - SAN Backup Solutions

Author

SAN Backup Solutions

dan_s_smith@hotmail.com

2004-08-24, 2:51 am

Hi,

I have joined a site with a currently very complex backup setup with 3
HP SAN's (2 HSG's 1 EVA). They are currently using business copy to
snapshot drives onto the main backup server to backup using backupexec
across fibre. This is fine but along side this some 6 oracle servers
are also using backupexec installed on themselves to backup their
databases across the fibre to the same two 4 drive libraries at the
same time.

I have two issues with this - we are experiencing device i/o timeout
errors every now and then (I assume due to different streams from
different copies of backupexec sending to the tape drive units at the
same time?) which is causing jobs to fail.

The main problem however, is when I have to shut down one of the tape
librarys for whatever reason. In order to get the librarys back online
in backupexec I have to reboot all of the servers - not really possible
with live oracle systems (win2k boxes - restarting the Backupexec
Services doesn't seem to cut it).

I would like to move to netbackup 5.1 to solve these problems and
wondered what people thought of it? I have never used it personally but
as far as I understand it runs with a "media server" and just an agent
on each machine so I won't run into the having to reboot live servers
to get drives online issue and also of course one central management
point rather than having to set jobs up and manage several different
servers at the same time?

Another thing - I believe it can also backup the same data to two
drives at the same time? (we are using one library for offsite jobs and
one for on so that would be helpful too)
Any thoughts or comments would be appreciated.

Thanks

Dan

Faeandar

2004-08-24, 6:53 pm

On 24 Aug 2004 01:53:50 -0700, dan_s_smith@hotmail.com wrote:

>Hi,
>
>I have joined a site with a currently very complex backup setup with 3
>HP SAN's (2 HSG's 1 EVA). They are currently using business copy to
>snapshot drives onto the main backup server to backup using backupexec
>across fibre. This is fine but along side this some 6 oracle servers
>are also using backupexec installed on themselves to backup their
>databases across the fibre to the same two 4 drive libraries at the
>same time.
>
>I have two issues with this - we are experiencing device i/o timeout
>errors every now and then (I assume due to different streams from
>different copies of backupexec sending to the tape drive units at the
>same time?) which is causing jobs to fail.
>
>The main problem however, is when I have to shut down one of the tape
>librarys for whatever reason. In order to get the librarys back online
>in backupexec I have to reboot all of the servers - not really possible
>with live oracle systems (win2k boxes - restarting the Backupexec
>Services doesn't seem to cut it).
>
>I would like to move to netbackup 5.1 to solve these problems and
>wondered what people thought of it? I have never used it personally but
>as far as I understand it runs with a "media server" and just an agent
>on each machine so I won't run into the having to reboot live servers
>to get drives online issue and also of course one central management
>point rather than having to set jobs up and manage several different
>servers at the same time?
>
>Another thing - I believe it can also backup the same data to two
>drives at the same time? (we are using one library for offsite jobs and
>one for on so that would be helpful too)
>Any thoughts or comments would be appreciated.
>
>Thanks
>
>Dan

One thing I will suggest is to not run a media server on the Oracle
servers themselves. Either back them up over the wire or use SSO for
SAN option. There are quite a few things that can dork up a media
server and require a reboot, which in this case would again be your
Oracle servers.

I've heard SSO can be flaky and even more so the SAN SSO option,
though I've never used either. Over the wire is definitely an option
in most cases. We backup a few 500gb databases over the wire as well
as the myriad 50-100gb databases, it all depends on your network usage
for data.
If that's a problem then maybe a dedicated backup network, though most
people really don't need it if they dig into their data traffic
patterns.

Also, given the type of array's you mentioned it should not be too bad
to setup another mirror config. Mirror the db's to another LUN, break
that off and mount it directly on the backup server.

Just some thoughts.

~F

dan_s_smith@hotmail.com

2004-08-27, 5:46 pm

Thanks for your reply. Yup the whole idea of going to netbackup is to
remove the media server part from the oracle machines. I have a box in
mind for a media server but what should I be concentrating on when
speccing this? Fast Processors, Large RAM, Fast disks, etc? I have a
dual Xeon with a couple of gigs in it so I reckon that should be fast
enough?

I would agree SSO is a bit flaky (esp with these timeout errors) so
another reason for moving. Most of the databases are in the 80-300gig
category.

With regards to the mirroring, I already do this with some disks but
with all these other servers being incorporated will I not run out of
LUN numbers on the media server (each oracle server has 4 separate
disks that need to be mirrored and backed up)? So I will need to mount
a couple at a time or can I put more hba's in the media server and
mount to separate hbas?

Faeandar

2004-08-27, 5:46 pm

On 27 Aug 2004 02:32:09 -0700, dan_s_smith@hotmail.com wrote:

>Thanks for your reply. Yup the whole idea of going to netbackup is to
>remove the media server part from the oracle machines. I have a box in
>mind for a media server but what should I be concentrating on when
>speccing this? Fast Processors, Large RAM, Fast disks, etc? I have a
>dual Xeon with a couple of gigs in it so I reckon that should be fast
>enough?

I have been out of the backup role for a while now but iirc it's all
about cpu and IO. You want to push/pull data as fast as possible over
the medium. NBU doesn't really cache anything to tape but the disk
stu's may be a different story.

>
>I would agree SSO is a bit flaky (esp with these timeout errors) so
>another reason for moving. Most of the databases are in the 80-300gig
>category.
>
>With regards to the mirroring, I already do this with some disks but
>with all these other servers being incorporated will I not run out of
>LUN numbers on the media server (each oracle server has 4 separate
>disks that need to be mirrored and backed up)? So I will need to mount
>a couple at a time or can I put more hba's in the media server and
>mount to separate hbas?

You would only mount whatever you're backing up at the time, not all
of them all the time. So stagger the schedule a bit, probably not
difficult since you can only backup so much data at any one time, and
swap out the LUN's as completed.

~F