NDMP backup applications

[Dan Pritts]

Hi folks -

We currently use veritas netbackup to back up our cluster to tape via
NDMP. Works fine.

We have been considering moving to cloud storage for our offsite
backups. Maybe for all our backups.

However, Veritas's pricing for their S3 option is a lot more than we
want to pay. $3k per terabyte protected, $500/yr thereafter. That pays
for a lot of tapes and offsite trips.

Isilon's supports netbackup, emc networker (of course), and commvault.
I haven't gotten a quote from commvault but the web site suggests that
they are pricey.

ZManda supports Isilon with their Enterprise Amanda. Anyone tried it?

Any other suggestions?

thanks
danno
--
Dan Pritts
ICPSR Computing & Network Services
University of Michigan
+1 (734) 615-9529

[Jerry Uanino]

How much data are we talking about?

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Dan Pritts]

33TB in current backup set.

Current backup schedule is:

Full, 13mo retention, quarterly
Cumulative incr, 13mo retention, monthly
incr, 6mo retention, weekly
supplemented by nightly Isilon snapshots

Cumulative size of current backups is ~200TB


this uses 111 LTO4 tapes, so about 89TB after LTO compression.   Some of the data is very compressible, ASCII format sas data files & the like.  And yes, we are thinking about how to replace the LTO4 infrastructure, and that is comingled with this cloud backup conversation. 

I've considered buying a big chunk of SATA disk an using it as a "basic disk" storage unit in netbackup, and uploading to glacier from there.  Seagate "archive" SMR disks are only $225/8TB.  Seems kind of kludgy, although the combination of that and glacier could arguably replace my current tape infrastructure for not a whole lot of cash.

As an aside, Oracle is selling a glacier-like cloud storage for $.001/GB/mo.  For comparison, Glacier is $.007/GB/mo.  None of the  backup vendors I've looked at supports the oracle product natively, though.  S3, sometimes Google, less often Azure.  Which makes sense from a market penetration POV. 

The annual subscription list price for Amanda enterprise with all the options I'd want is about half of what I currently pay for annual maintenance on netbackup.  $3000 I think. 

Minor correction to what I wrote previously.  According to the "backup & recovery guide" for 7.2, EMC supports the following  for NDMP. 

Symantec NetBackup
EMC NetWorker
EMC Avamar
Symantec Backup Exec
IBM Tivoli Storage Manager
Dell NetVault
CommVault Simpana
ASG-Time Navigator

Not all of these have "real" cloud support.  e.g., NetVault supports Amazon's S3-backed VTL but no native S3 support. 

danno

Jerry Uanino

March 16, 2016 at 6:04 PM

How much data are we talking about?

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Dan Pritts

March 16, 2016 at 5:24 PM

Hi folks -

We currently use veritas netbackup to back up our cluster to tape via NDMP. Works fine.

We have been considering moving to cloud storage for our offsite backups.  Maybe for all our backups.

However, Veritas's pricing for their S3 option is a lot more than we want to pay.  $3k per terabyte protected, $500/yr thereafter.  That pays for a lot of tapes and offsite trips.

Isilon's supports netbackup, emc networker (of course), and commvault.  I haven't gotten a quote from commvault but the web site suggests that they are pricey.

ZManda supports Isilon with their Enterprise Amanda.  Anyone tried it?

Any other suggestions?

thanks
danno

[Jerry Uanino]

So..... what we did here was, we kinda got sick of netbackup.

We started augmenting with snapshots just to save tape, then we pulled the trigger on an "archive cluster" for us.

1PB and just stop using your expensive/annoying backup software.  Setup syncIQ jobs and snap on the target.

You can calculate some savings because you're not snapshotting on your expensive storage anymore, snap on the target. Then turn on dedup on the target because it's cheap to keep snapshots there. Get the largest cheapest NL nodes.

We decided the failure rate of tapes, backup software, upgrades, etc wasn't worth it.

We dump some other "file based backups" too there like sql dumps that the SQL server guys just configure in sql server. 

Some giant tarballs people generate, etc.  Anything not in production runstream.  Half this stuff was ending up on tape and people were getting "creative" on managing their retentions.  Now we just control it with snapshots. Wanna see what is in backups. Give them a mount, ls. Capture the snapshot listing output and expose it.  Much easier.

We haven't dumped netbackup yet, but NDMP was just too slow anyway and always seemed to fail for some reason (buggy vendors).

SyncIQ does seem to be solid and worth it so far.

Your 33TB backup set is a little big for an Isilon SD edge, but I've considered the scenario where small backup sets could go to Isilon SD edge running on vmware. (36TB max cluster size right now).

I think the day of backup software is going to end soon, you can't be slurping 33TB around easily.  Would you ever want to restore from that, or my 300TB cluster. No way. You'd be fired before you finish.

Advantage of snapshots too is you can just share out the export/cifs share in a disaster and mount it.  Done.

Yes, it costs a lot of money. Add up ditching your tape library, netbackup, ndmp license, the by TB license you would need for veritas dedupe, etc etc.

Don't forget sys admin costs for maintaining that netbackup setup and upgrading it every year.  Oh and if you can manage to ditch a media server or two in the process you no longer need to refresh that hardware few years.

I'm lucky, we had the need for other archive uses, but I piggybacked on that purchase to reduce reliance on NetBackup.

Do you really like using a java interface from 1998? Didn't think so.

Map out a price, if you do I would be curious if the headache is worth it. I love knowing I can just mount it if I need to somewhere. No restore needed (or reverse your syncq).

p.s. SD Edge comes with SyncIQ and all licenses in the purchase price. Untested by me, but seems interesting (will test soon).

[Donald King]

We did the DR Isilon cluster thing too.  We have about 200TB on Isilon.  NBU using NDMP was always a pain but at that data size it was becoming unusable (3 -4 days to get a full backup).  We have two locations with a fast private network connection between them.  So we bought a second Isilon cluster, put it at the other location, and replicate snapshots nightly.  We sized the backup cluster HUGE to keep 1 year of snapshots.  All NL nodes of course in the DR cluster.  

No more tape, backups are off site every night.  Of course it was a larger capital expenditure to get the DR cluster - but well worth it in the long run.  We now have reliable nightly backups - not the "well maybe NDMP will work this week" backups.  

-Don

[Donald King]

Oh - I forgot to mention the downside to the DR cluster approach.  You still have to keep  your backup accelerator in your production cluster for as long as you have NDMP tape images that are not expired.  We keep some backups for 5 years, so my NBU / NDMP / backup accelerator has to stay around for a while in case those images need restored.

-Don

[Dan Pritts]

The big downside for us is that we don't have a suitable offsite location to put a DR cluster.   We could probably arrange something at a regional campus of our University, but the furthest one is only 55 miles away.   not as far as I'd like. 

We could of course arrange other hosting, but it's a hassle and a significant expense compared to cloud storage like Glacier. 

Bottom line, NDMP is manageable for us up to this point, and our data growth isn't that crazy.  I want to keep doing it, but switch out tape for cloud storage...but jeez, the netbackup licenses are expensive.  Probably more, in fact, than we'd pay for a DR cluster, now that I think about it.  Huh. 

w/r/t the backup accelerator, I thought that we could do restores through the backup server - any node can run the ndmp service, the accelerator is just there to talk to tape drives.    So as long as netbackup is connected to the drives i should be OK.   I'm 99% sure that the ndmp service runs on all nodes, and you can't restrict it to the accelerator - I wanted to do that for security reasons. 

danno

Donald King

March 21, 2016 at 9:22 AM

Oh - I forgot to mention the downside to the DR cluster approach.  You still have to keep  your backup accelerator in your production cluster for as long as you have NDMP tape images that are not expired.  We keep some backups for 5 years, so my NBU / NDMP / backup accelerator has to stay around for a while in case those images need restored.

-Don

On Monday, March 21, 2016 at 9:20:31 AM UTC-4, Donald King wrote:

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Jerry Uanino

March 17, 2016 at 6:37 PM

So..... what we did here was, we kinda got sick of netbackup.

We started augmenting with snapshots just to save tape, then we pulled the trigger on an "archive cluster" for us.

1PB and just stop using your expensive/annoying backup software.  Setup syncIQ jobs and snap on the target.

You can calculate some savings because you're not snapshotting on your expensive storage anymore, snap on the target. Then turn on dedup on the target because it's cheap to keep snapshots there. Get the largest cheapest NL nodes.

We decided the failure rate of tapes, backup software, upgrades, etc wasn't worth it.

We dump some other "file based backups" too there like sql dumps that the SQL server guys just configure in sql server. 

Some giant tarballs people generate, etc.  Anything not in production runstream.  Half this stuff was ending up on tape and people were getting "creative" on managing their retentions.  Now we just control it with snapshots. Wanna see what is in backups. Give them a mount, ls. Capture the snapshot listing output and expose it.  Much easier.

We haven't dumped netbackup yet, but NDMP was just too slow anyway and always seemed to fail for some reason (buggy vendors).

SyncIQ does seem to be solid and worth it so far.

Your 33TB backup set is a little big for an Isilon SD edge, but I've considered the scenario where small backup sets could go to Isilon SD edge running on vmware. (36TB max cluster size right now).

I think the day of backup software is going to end soon, you can't be slurping 33TB around easily.  Would you ever want to restore from that, or my 300TB cluster. No way. You'd be fired before you finish.

Advantage of snapshots too is you can just share out the export/cifs share in a disaster and mount it.  Done.

Yes, it costs a lot of money. Add up ditching your tape library, netbackup, ndmp license, the by TB license you would need for veritas dedupe, etc etc.

Don't forget sys admin costs for maintaining that netbackup setup and upgrading it every year.  Oh and if you can manage to ditch a media server or two in the process you no longer need to refresh that hardware few years.

I'm lucky, we had the need for other archive uses, but I piggybacked on that purchase to reduce reliance on NetBackup.

Do you really like using a java interface from 1998? Didn't think so.

Map out a price, if you do I would be curious if the headache is worth it. I love knowing I can just mount it if I need to somewhere. No restore needed (or reverse your syncq).

p.s. SD Edge comes with SyncIQ and all licenses in the purchase price. Untested by me, but seems interesting (will test soon).

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Dan Pritts

March 17, 2016 at 12:28 PM

[Brad Schlicht]

We previously used Networker and the NDMP accelerator to backup our production Isilon cluster back when we only had around 65 TB of data. The backups were generally take around 4 days and would fail here and there. As our data continued to grow, I moved to the snap and replicate strategy. We have snapshot schedules on our production cluster and replicated to our DR cluster, which also has snapshot schedules. However, we still have a requirement to send a small subset of our Isilon data of-site to Iron Mountain. So I kept Networker around and use it to backup (~20TB of data) to a Quantum tape library. This runs every two weeks and takes about 14-15 hours to complete.