Backup Speeds

[dwilson]

We are in the process of upgrading our server to NetBackup 7.1. While doing this we are upgrading our server to a newer more robust hardware platform. I have followed the procedures in bringing our current 6.5.6 environment over to the new server (same server name, install directory, cataloge, etc.), and have everything up and running without any errors. My problem lies with the backup speeds that we are seeing after the migration. Policies that were running fine on the older production environment are now failing, and not finishing in there alotted schedule windows. I have compared the Activity Monitors of the two and can see the differences between the two speeds (new server is half the speed). Does anyone know of any tweaks or reasoning that this may be happening. It seems on our previous server setup there were some things I could do to help out this process, but I can not recall or located what it was. I know it cannot be the server specs, but I will include them any ways.

Old: Dell Power Edge 2850 (3.2 GHZ - single, 4GB RAM, 32-bit Server 2003)

New: HP BL45P (2.2GHZ  -quad, 29GB RAM, 64-bit Server 2008 R2)

[RonCaplinger]

I'd suggest the normal procedures for performance tuning in the "Veritas NetBackup Backup Planning and Performance Tuning Guide".  Check NIC speed/duplex settings first, but taht's probably not the problem.

If you moved everything from one box to the next, you may not have set up the NET_BUFFER_SZ the same as before.  Also run the tuning steps in the guide between the new server and the clients.  And if your backups are going directly to tape, you should check your buffer size and number of buffers.  All this would be found in the Tuning Guide.

[Marianne]

See if this TN helps: http://www.symantec.com/docs/TECH145791

[pdsimpson]

I also saw slower backups when I switched to an HP C-Class Blade Server. In addition to things like NET_BUFFER_SZ, you should check the firmware levels on your blade & the blade enclosure itself.

Also, are you teaming the NIC's on the blade? Have seen that impact performance also, runs best with single-NIC.

Also look at what other systems are in that blade enclosure & the amount of data they each transfer... could be overwhelming the blade chassis.

Lastly, is your NBU DB on local or SAN disk? If on SAN, did you split the app / logs / DB onto 3 different LUNS? All on the same LUN WILL impact performance, best practice is on local disk, but if on SAN, split the LUNS.

[dwilson]

Alright, thanks for all the replies. I ended up openning a case # with Symantec and after looking through the bptm logs (lots of waited for empty buffer XXXXX times errors) it seems that we are experiencing an issue with either our tape drivers or maybe our fiber HBA. I am going to upgrade firmware and drivers today and then run some more test to see if that resolves the issue. Will follow up afterwards.