Video Screencast Help
Search Video Help Close Back
to help
Not able to make it to Vision this year? Get a sampling in the Best of Vision on Demand group.

Poor duplication performance

Created: 07 Jun 2010 | 8 comments
AlGon's picture
0 0 Votes
Login to vote

Netbackup 6.5.3.1

Windows 2003 Server

Hello

We are using Netbackup vault to duplicate weekly and monthly full backups from our VTL (Quantum DXI7500 Inbuilt De-Dupe) to an LTO4 tape library (Quantum PX500).
We’re duplicating around 5.5TB of full backup data which currently takes around 5 days to complete.

The media server that handles this process is an HP Proliant DL380 G5 with two 2.66 GHz dual core processors and 4GB of RAM.  This server is fibre attached (4Gbs) to the VTL and LT04 tape library.

The original backups to the VTL are not multiplexed.

It seems that when the duplication starts its reasonably quick but then decreases in speed over time.  Looking at the performance monitor on the VTL I noticed the tape read performance starts around 60MB/S then gradually degrades to around 20MB/S after which it fluctuates between 0MB/S and 20MB/S.  It is as if the tape it is writing to is stopping/rewinding/stopping/rewinding.

We increased the buffer, which made a significant increase in speed for the first couple of hours and was a consistent 60-70MB/S.  However, when we left the dupe job running over the weekend, we found it had gone back to the start-stop behaviour like it had before.
Current buffer settings are:

NUMBER_DATA_BUFFERS = 32
SIZE_DATA_BUFFERS = 262144

Any help will be greatly appreciated.

Thanks

Comments

Reagan's picture
07
Jun
2010
1 Vote +1
Login to vote

Similar situation

Hi,

Duplicating from Quantum DXi7500 to a Quantum tape library with 2 LTO-4 tape drives.

Found these settings to work the best under most conditions.

NUMBER_DATA_BUFFERS = 64
SIZE_DATA_BUFFERS = 262144

Hard setting the speeds of the HBA's and the tape drives to 4Gb/s may make a difference.

If possible, try using separate HBA's for the DXi and the tape library.

Hope this helps.

AlGon's picture
07
Jun
2010
0 Votes 0
Login to vote

Hi Reagan, thanks for the

Hi Reagan,

thanks for the quick reply.  Did you encounter similar problems?

Not possible to use a separate HBA at present but will try the buffer settings you mentioned.

Thanks

Ed Wilts's picture
07
Jun
2010
0 Votes 0
Login to vote

Sounds familiar...

We had a long-standing call with Symantec on DSSU destages (which are duplication jobs).  We were typically seeing about 20MB/sec, about the same as you're seeing.

Symantec was unable to find the bottleneck in our environment even though we had engineers on site and our call escalated to product management.   They said they couldn't find any issues in the code though.  Based on what you're seeing, I'm not convinced that the code doesn't have an artificial limit in there...  We finally gave up and very few jobs use DSSUs now and our throughput has significantly increased.

If this post helps you, please add a vote.

If this post answers your question, please mark it as a solution.  Thanks!
 

Reagan's picture
07
Jun
2010
0 Votes 0
Login to vote

encountered similar problem

The buffer settings are the only tuning settings with NetBackup that can help out with duplications. 
Tried numerous buffer settings, but the ones previously mentioned worked the best.
A single HBA is being used for both the VTL and the Tape library.  The tape library is connected to a 2Gbps switch, which I believe is currently the bottleneck.  The DXi is directly connected to the master server.
The overhead of rehydrating data from the Quantum DXi to tape is around 10% - 15%.  Even with that, my duplication rates were lower than expected.  Symantec support was unable to find anything wrong with the configurations and pointed the problem to hardware.  

 

dubplate's picture
07
Jun
2010
0 Votes 0
Login to vote

We had similar issues (but

We had similar issues (but with a quite different setup) that seem to finally be resolved after many months of troubleshooting by upgrading to NBU 6.5.5. We went straight from 6.0 MP7 to 6.5.5 though as a latch ditch effort after trying all sorts of things, so it could have been bugs/fixes in any 6.5.x release in between 6.0 MP7 and 6.5.5. 6.5.3.x seems to have the most vault related fixes according to the release notes but I see that you are already on that.

AlGon's picture
30
Jun
2010
0 Votes 0
Login to vote

We’ve tried various

We’ve tried various buffer settings and have improved performance slightly but not by a significant amount.
In the end I worked out how to create a second partition on the VTL without de-duplication enabled.
I tested duplicating from this partition to tape and the transfer was much quicker and more like what I would expect going from a VTL to LTO4 tape.

The VTL is only 9TB in size. The amount of weekly full backup data sent to the VTL is currently around 5 Terabytes and with the de-duplication turned on it allows us to retain copies on the VTL for a good number of weeks.  If I turn de-duplication off I’ll only be able to retain full backups for about a week and incremental backups for possibly two.
 
I have a few ideas to get around this and would like some advice.
One idea is to set up a storage life cycle policy for our weekly and monthly full backups.  This would be configured to send one copy to the VTL (de-duplication turned on and retain for a number of weeks) and another copy directly to the physical tape library for offsite DR and long term storage.  The incremental backups would also go to the VTL as normal.  Therefore I would no longer be using vault to handle duplication to tape.
 
What would be the main drawbacks of not using vault?

Are storage lifecycle policies a lot better in version 7 than 6?  (We will soon be going to 7)

I’m not very familiar with storage life cycle policies and would like to know how they compare to using vault.

Another idea would be to create two partitions on the VTL, one with de-duplication enabled and the other without.  Weekly and monthly full backups would go to the non-de-duplicated partition (only retain for a week due to the small amount of space available on the VTL) and send the incremental backups to the de-duplicated partition (utilizing at least some of the de-duplication technology on the VTL).  I could then use vault to duplicate the data from the non-de-duplicated partition on the VTL directly to physical tape.

Thanks again.

wr's picture
30
Jun
2010
0 Votes 0
Login to vote

De-dup De-dup De-dup

De-du-du-du-du  De-dup  indecision

Sorry I have no quick answer, rather another question: the slowdown is the VTL un-de-duping?  (Is that even a word?)  It seems each new technology has its quirks and I'm trying to sort them all out too. 

good Will backing-up

AlGon's picture
01
Jul
2010
0 Votes 0
Login to vote

I guess my main questions

I guess my main questions are

what would be the main drawbacks of not using vault if i decide to use storage life cycle policies instead?
 
Are storage lifecycle policies a lot better in version 7 than 6?  (We will soon be going to 7)

Thanks again.