> I have a file server with approximately 10 drives, 500 GB each. 5TB to
> back
> up. It's your run-of-the-mill Win2K3 file server for all our users. SAN
> attached of course.
>
> We back up to LTO2.
>
> Any ideas on how to better back this up? A full backup takes almost two
> days. We've moved to the advanced client for snapshot backups, and we
> have
> multiple streams in the policy.
>
> Of course if the backup fails, we have to re-run the whole policy, which
> either means manually cancelling parts/streams, or wasting time/tape
> backing
> up the same data.
>
> Would it be a better idea to use synthetic fulls? Or look into a product
> like the Enterprise Vault?
>
> If we back up one drive for a full backup we can reach speeds of 25 MB/sec
> to tape.
>
Need a little more info

How many partitions?
Are there restrictions on backup hours?
How many LTO drives do you have?
You mentioned SAN attached storage, is this a media server?
Are you doing LAN or SAN based backups?
How many LAN or SAN interfaces are you using?

Paul W.

>How many partitions?

I think I said 10 drives, I'm not sure what you're getting at here. Each
drive is a separate lun/array on the same SAN storage unit, if that's what
you want to know. But the win2K3 server sees them as 10 distinct volumes/luns.

>Are there restrictions on backup hours?

- ideally we're talking fulls over the weekend simply to avoid any performance
impact, but not practically. We are *making* the windows its just that it
takes a long time, and is prone to failure.

>How many LTO drives do you have?

- a total of 6

>You mentioned SAN attached storage, is this a media server?

This is a windows 2003 file server; it is not the media server (but it could
be if it were the right thing).

>Are you doing LAN or SAN based backups?

- dedicated gigabit lan;

>How many LAN or SAN interfaces are you using?

- 1 private lan as far as backups go


I don't suspect any network issues, we've gone through quite a bit of diagnosis.
I can back up exchange (a large stream) at 40-50 MB/sec from the same san.
This is a multi-million-file file server, so it has the typical problems
you've seen discussed on the newsgroup before. Moving to snapshot backups
did improve performance. If I back up one drive I typically get about 25
MB/sec. If I back up 5 drives simultaneously (to the same target) I get
a rough aggregate of maybe 35 MB/sec.

We'd be able to move to san-based backups I guess, but would need to deploy
some infrastructure - tape drives aren't fiber-attached for example. And
I'm not clear that just upping the performance a bit is the answer. It takes
say 36 hours to back up all the data (assuming no failures). If I improve
my performance by 100% its still going to take 18 hours - and the failures,
which occur, still take a re-run.

What tunable parameters have you applied for your San Attached drives / OS
?

> I think I said 10 drives, I'm not sure what you're getting at here. Each
> drive is a separate lun/array on the same SAN storage unit, if that's what
> you want to know. But the win2K3 server sees them as 10 distinct
> volumes/luns.
Sorry, you said drives, which means to me individual disk platters. So these
are LUNs of the array. Are these LUNs all RAID5? Are they carved out of the
same or mulitple RAID groups? What is the array on the backend? How much
cache does this array have? What other resources are utilizing this array at
the same time? Also, is 500 GB the size of each LUN or is it the size of the
data set on each LUN?

> - ideally we're talking fulls over the weekend simply to avoid any
> performance
> impact, but not practically. We are *making* the windows its just that it
> takes a long time, and is prone to failure.
Your statement 'prone to failure' causes me some concerns as this seems to
indicate to me that you encounter failed backups at a regular pace. What is
the percentage of failed backups that you are experiencing? Also, it sounds
like you need to make your window bigger and there are three ways to do
that:

1) Add more drives giving you more available drive-hours.
2) Reduce the size of the data. (yeah, riiiiiiiiiiiiiight)
3) Increase the hours available to perform backups.

In today's environment, keeping weekends as the only time to perform all
your Full backups is no longer practical due to increase in data set sizes.

My recommendation, without looking at any tunable parameters, would be to
create 10 policies (or even 5 policies of 2 LUNs each), each policy would
backup one LUN. Then I would spread these policies out thruout the week.
If the retention parameters are the same, then this will not greatly
increase your tape usage. Also, this will afford you greater flexibility to
respond to single failed backup jobs. Downside, is that it will increase
your administrative complexity due to the increased number of policies and
the distributed nature of the full backups. Or you need to have someone
available to address failures as they occur instead of waiting until Monday
AM.

>
>>How many LTO drives do you have?
>
> - a total of 6
That may not be enough, I'm also assuming that you have other backup targets
also? In addition to duplication, vault, restores, etc?

>
>>You mentioned SAN attached storage, is this a media server?
>
> This is a windows 2003 file server; it is not the media server (but it
> could
> be if it were the right thing).
>
Sounds like it wouldn't be as your tape drives are not SAN attached.

>>Are you doing LAN or SAN based backups?
>
> - dedicated gigabit lan;
>
I'm assuming that you have taken a look at switchport statistics during peak
demand?

>>How many LAN or SAN interfaces are you using?
>
> - 1 private lan as far as backups go
If you are running at Gig, that should be plenty.

>
>
> I don't suspect any network issues, we've gone through quite a bit of
> diagnosis.
> I can back up exchange (a large stream) at 40-50 MB/sec from the same san.
> This is a multi-million-file file server, so it has the typical problems
> you've seen discussed on the newsgroup before. Moving to snapshot backups
> did improve performance. If I back up one drive I typically get about 25
> MB/sec. If I back up 5 drives simultaneously (to the same target) I get
> a rough aggregate of maybe 35 MB/sec.
That performance sounds about right for your configuration, given the small
file performance hit inherent in file data backups.


Paul W.

I'm not sure what you're asking here - for the disk drives on the servers
themselves, or on the tape drives (scsi attached), etc?

"Mike Culbertson" wrote:
>
>What tunable parameters have you applied for your San Attached drives /
OS
>?
>
>

Your best bet is to implement Enterprise Vault (Filesystem Archiving). Since
you're backing up a lot of data that doesn't change, you can archive the
data once, and backup it *once*, and you're done. Once you close out a vault
store partition, it is guaranteed not to change. All your subsequent backups
will contain new data that hasn't been archived.

If you want to stay exclusively with NetBackup, here are a few things to
note:
(1) You said SAN attached, but is the backup traffic going of FC or ethernet?
(2) If it is over FC, then your fileserver is a "NetBackup SAN Media Server",
correct? If not, that's a big step in the right direction.
(3) Buy LTO3 drives. The limitation will still be your server, but if you
increase the write and read buffers, you can get more going to tape.
(4) You're either using a large internal backplane or an FC JBOD, correct?
Buy a storage array with NDMP and additional FC ports for the tape drive(s).
When looking at a storage array and a server in the same tier, the storage
array can always stream faster than a server, so use this advantage to dump
direct to tape from the array. This is how you get about 125MB/s to tape
or higher (it all depends on the array of course).

Hope this helps!

Best Regards,
Nick Ross

"Mike Robinson" wrote:
>
>I have a file server with approximately 10 drives, 500 GB each. 5TB to
back
>up. It's your run-of-the-mill Win2K3 file server for all our users. SAN
>attached of course.
>
>We back up to LTO2.
>
>Any ideas on how to better back this up? A full backup takes almost two
>days. We've moved to the advanced client for snapshot backups, and we have
>multiple streams in the policy.
>
>Of course if the backup fails, we have to re-run the whole policy, which
>either means manually cancelling parts/streams, or wasting time/tape backing
>up the same data.
>
>Would it be a better idea to use synthetic fulls? Or look into a product
>like the Enterprise Vault?
>
>If we back up one drive for a full backup we can reach speeds of 25 MB/sec
>to tape.
>
>