Drive assigment

[Bobski]

Hi,

I have a question (probably a silly one) regarding drives - we have 3 lto6 drives housed in LS150 Oracle robot. We have one master server with NetBackup 7.6.1.We have one STU. We have 2 sets of full backups at the weekend, one is for offsite storage.In this configuration when one job is running it is using all 3 drives thus causing the other job (offsite one) to queue until there are available drives. This isn't ideal as the backups are quite slow so the backup window overruns and the jobs waiting for available drives are then failing.

Is there a way of making sure that, lets say 2 of the drives are dealing with one job (in house set) and the other drive is dedicated to another job (offsite set) ?

Thanks

[areznik]

Is there a way of making sure that, lets say 2 of the drives are dealing with one job (in house set) and the other drive is dedicated to another job (offsite set) ?

Yes, you can set up 2 new STUs for the same media server/library; One with Concurrent write drives = 2 and the other with concurrent write drives = 1; then point policies at the different storage units to separate your backups. You can use schedule storage overrides if you only want to do this for certain backup types (weekend fulls for example). 

(!) However, this is not an efficient use of your tape drives. If Job 2 is finished Job 1 will continue to run on its dedicated 2 drives even though now a third is available. Depending on what kind of backups you are doing you might be able to set up multiplexing and allow both jobs to write to all 3 drives simultaneously. If you're interested in that, let us know what kind of policy types these jobs are and if they are using the same retention levels and volume pools. 

[sdo]

Or keep the one storage unit, use multiplexing, use one retention level, but use priorities on policies so that important jobs multiplex in sooner.

[Marianne]

My vote is also for multiplexing. Unless current is maxing out drive speed and running at sustained max possible throughput. Experience has taught us that this is hardly ever seen. Mpx of 4 in stu and schedules as well as 2-4 jobs per client is normally a good starting point.

[Marianne]

Just wondering about the 2 sets of backups... Same data backed up twice? Only difference longer retention and probably different pool? If so, why not config SLP with backup to onsite pool and retention and duplication to offsite pool and retention?

[Bobski]

Yes, it is same data with different retention levels and different pools. I have just briefly looked over SLP which I  understand I could use to schedule  duplication of backups. Does SLP minitor when the jobs are finished and then start duplication or is that not how it works ?

Also is this fairly easy to setup ? 

[Bobski]

I've just had a look to setting up new SLP policy, however why do I only see Backup and Import in Operation field ? Shouldn't I also see Duplicate ?

I've attached the screen shot.

[sdo]

Click "Add" then choose the operation type.

[Bobski]

I think I've figure it out (by properly following documentation, duh ! )

Best thing I ever did - joining this forum :)

Lets see how this goes.

Thanks all for your help so far and thank you Marianne for pointing out SLP option.

[Marianne]

Well done for doing own research and reading documentation!

About this question:

Does SLP monitor when the jobs are finished and then start duplication or is that not how it works ?

Yes. That is the default behaviour.

If you configure SLP windows, the duplication will only take place during this period.

Because you have only 3 drives, we can safely assume that SLP duplications will use 1 drive to read and another to write.
If you have configured suitable buffer sizes, the duplication should go quite fast as no network is involved.

You may want to look for SLP Best Practice Guide for more hints and tips.

[Bobski]

One more question, bearing in mind I have 3 drives in one STU, can I just use this one STU in my SPL policy or should I create new STU (lets say I'll call it Duplication) with one drive and configure the SLP's secondary operation (duplication) with that, so the SLP config might look like in the file attached....

[Marianne]

Use the same STU for backup and duplication.

NBU will assign drive(s) to read and write for duplication step.

[Bobski]

Hi ,

Just an update on my SLP adventure. I've created one SLP policy and assigned it to my clients. All was working good but unfortunatelly I have created new window for secondary job (duplication) and I stupidly scheduled to finish by mid tuesday. That wasn't a good move because our backups are very slow so by the time they've finished the duplication windows has closed and few server have not been done.

Now, I've learned from my mistake and when I created second slp policy I've left the default24x7 window. However for some reason this second slp didn't start first (even though I gave it 999) and instead the normal policy went ahead and I could see slp queueing.

Any idea why the slp , month_end wouldn't start even though it had highest priority ?

I've attached the screen shots of both SLPs, ( I've disabled them for a moment) , also at the very end is the screen shot of speed I get on our main file server - that's way too slow for LTO6's , right ?

Thanks

[Marianne]

If you refer to 'normal policy' as backup policy, then this seems right.
If memory serves me right, backup has priority over duplications.

Have a look in Host Properties -> Master -> Default Job Priorities

It seems you probably do not have enough tape drives... or not performing optimally (as you have noticed).

I see you have compression enabled in the 1st policy screenshot. 
Why is that?

Compression on client is resource intensive and will slow down backups.
Client compression is only advised for clients on very slow networks.

In the 2nd screenshot you have 'Follow NFS' enabled.
NFS backups will also be slow as data transfer will be doing 2 network hops.

You need to investigate each component in the data path - the biggest culprit is normally slow read speed from client disk.

If you are getting fast enough to the media server, then look at increasing SIZE_DATA_BUFFER and NUMBER_DATA_BUFFER.

Do you have multiplexing and multistreaming enabled?
4 jobs per drive @ 15 - 20 MB/sec will certainly give you much better throughput.
We have seen that MPX of 4 gives good backup performance without impacting restore times.

[sdo]

If any drive has multi-plexing AND free notional job/usage count, then any pending lower priority backups will be allowed to join the multi-pled job set irrespective of pending higher priority backup jobs. I.e the multi-plex job possibilities take precedence.

[Mark_Solutions]

I have joined this conversation quite late but in relation to having 3 tape drives all getting used by one backups set which was your original question ... which can waste tapes .. a good way around that is, assuming you are using 2 different volume pools for each set, to limit the maximum number of partially full media. If that is limited to 2 then not only will it only use 2 tape drives but it will also only use 2 tapes at a time and not pick a third one until one of those two is full.

This works really well and i am currently using it for most of the volume pools in the environment i am working in.

A note on the SIZE_DATA_BUFFERS .. if you change that (shoudl be 262144 for LTO6 tapes) then you will need to relabel any used tapes (as long as they have expired of course!!) before they will take on the new block size.