Re: Bad performance Restore and Object Copy from StoreOnceDevices (666 Views)
Reply
Occasional Contributor
MatsH
Posts: 4
Registered: ‎12-16-2010
Message 1 of 5 (743 Views)

Bad performance Restore and Object Copy from StoreOnceDevices

Hi! Environment is DP 6.21 Have been testing around with StoreOnceDevices at a customer. We been testing backup, restore and object copy and seen som e strange results when comparing StoreOnce and traditional DP filelibrary. The data we're backuping up in test are files on one volume around 50 GB and WMware snapshots 20 - 50 GB each. Filelibrary and StoreOnce is in the same storagesystem both on RAID5 locally attached to the DP server. The DP server is the only Gateway to StoreOnceStore. In backup for both files and VMware the StoreOnce is around 5-10 % faster than the filelibrary In restores for both files and VMware the StoreOnce is around 20-30 % slower than the filelibrary, as low as 5-7 MB/s In restores for both files and VMware the StoreOnce is around 20-60 % slower than the filelibrary, as low as 1 MB/s Anyone seen this behavouir? Not sure where to start? //Mats
Trusted Contributor
jruffer
Posts: 184
Registered: ‎06-28-2011
Message 2 of 5 (740 Views)

Re: Bad performance Restore and Object Copy from StoreOnceDevices

Hi Mats,

 

I've been experimenting too and Session copy was a big problem.  With File Library using software compression it was taking 3-4 hours.  With StoreOnce it was taking 8-11 hours.

 

I had put that down to not using compression but maybe there's more.

 

Regards

 

Jeremy

Honored Contributor
André Beck
Posts: 515
Registered: ‎06-23-2005
Message 3 of 5 (726 Views)

Re: Bad performance Restore and Object Copy from StoreOnceDevices

[ Edited ]

Hi,

 

rehydrating (reading) from StoreOnce is expected to be slower than reading from a classic (non-DFMF) FileLibrary. The latter produces just large files as media containers, if your write parallelism isn't too bad, they will be allocated mostly contigous and thus read out with near sequential read throughput. Reading from SO is more of a random access pattern to a lot of small to medium sized files. That being said, I would not expect throughput to be as bad as you see it, unless the RAID backend's IOPS is really borderline: How many spindles of what RPM are there in the RAID5? Is the battery backup in good condition? What read throughput do you achieve on a small-file dominated file system on that RAID? What OS and choice of file system for the store? When you copy to a Null Device local to the gateway's media agent, is throughput still as bad? Is the RAID5 I/O-saturated when the data is reading? And last but certainly not least, is there enough available RAM in the box? SO really needs buffer cache to perform.

 

HTH,

Andre.

Honored Contributor
Tim Nelson
Posts: 3,727
Registered: ‎06-01-2000
Message 4 of 5 (697 Views)

Re: Bad performance Restore and Object Copy from StoreOnceDevices

Just chiming in for future reference.

 

I have been testing with a D2D4324.  Same issue/results.  The rehydration for either a restore or copy job is very very slow..800GB expected to take 10hours.

 

Scheduling a POC with another vendor.  We see if the results differ.

 

Honored Contributor
Sebastian.Koehler
Posts: 1,157
Registered: ‎02-27-2007
Message 5 of 5 (666 Views)

Re: Bad performance Restore and Object Copy from StoreOnceDevices

[ Edited ]

There is a defect in the RMA that causes this issue. The object copy is actually really quick, but it passes the data multiple times to the destination device. A fix for DP 6.21 and 7.00 is available and not included in the latest GA patches. Works like a charm for me.

Copy from StoreOnce device to diff device type takes more space than original data
http://support.openview.hp.com/selfsolve/document/KM1391228

Regards,
Sebastian

---
Assign a kudo to this post, if you find it useful.
The opinions expressed above are the personal opinions of the authors, not of HP. By using this site, you accept the Terms of Use and Rules of Participation.