UCS backup issue

[molerio]

We're trying to backup an oracle db using rman/tdp/lanfree.. the usual setup we have running elsewhere in our environment. it seem the backup takes 5xd longer when the database is hosted on the UCS.

We simplified the test to only run RMAN and we came up with the same results. backups run slower on the UCS then on other hardware.

Anyone experience this?

 

[moon-buddy]

What is your setup? Is the Oracle server on a blade server or is this a single UCS rack mounted unit?

We don't see issues with the latter setup.

 

[molerio]

UCS rack mounted unit.

we think its one of the settings on the virtual hba's that is causing the issue or the mix of nearline & sas storage its using.

 

[moon-buddy]

Virtual HBAs is a no-no for database servers unless you can allocate specific traffic on a particular virtual environment, i.e., public network only, backup network only, disk access only, etc. In other words, no mixing.

Use real Fiber and Network HBAs as much possible.

 

[molerio]

The vhba's settings were verified by UCS support.
Would it be more likely that the linux client (Linux x86-64,Client OS Level: 2.6.32-504.23.4.el6, TSM Client Version: Version 6, release 4, level 2.0), lannfree linux 5.5)
also needs a Hitachi driver for linux since it needs to access a VTL and its own storage (Hitachi) ?

 

[moon-buddy]

The vhba's settings were verified by UCS support.
Would it be more likely that the linux client (Linux x86-64,Client OS Level: 2.6.32-504.23.4.el6, TSM Client Version: Version 6, release 4, level 2.0), lannfree linux 5.5)
also needs a Hitachi driver for linux since it needs to access a VTL and its own storage (Hitachi) ?

It is possible that this is a driver issue but makes me wonder.

In your original post, you said it ran slower even when just running RMAN directly to the storage. This makes me think that there is either 1) a configuration issue, or 2) some driver issue.

Have you configured the Storage agent setup properly? TSM uses generic SCSI drivers for non-IBM tape drives (be it real or virtual). There is also options (in most VTL environments) to define virtual IBM Ultrium tape drives for which you will need IBM-type drivers.

 

[molerio]

storage agent is configured correctly. the VTL ( Protectier) is defined to use the LT03 enulation.
the linux client has the correct tape driver installed.

 

[moon-buddy]

S

storage agent is configured correctly. the VTL ( Protectier) is defined to use the LT03 enulation.
the linux client has the correct tape driver installed.

So if everything is correctly setup, then my conclusion is that the VHBA must be the issue.

Is there a way for your sysadmins to assign real HBA for this UCS as a test?