Directory Container Pool performance

[TykneeTym]

I am evaluating TSM at my current position as a replacement for NetBackup and DataDomain. I'm facing some resistance from people in other departments on this, so I'm doing an evaluation to prove that our current setup is horrible for the data set we have.

My IBM rep got me an evaluation OVF of 8.1. This wasn't how I wanted to test it, but apparently getting a demo version of the installer files was not happening (weird, but whatever). The virtual environment that we running this test instance in is not that great. The underlying disk is NetApp over iSCSI, which is something I would never solution. I'm not the storage engineer however, so I have to deal with what I have especially in the case of a evaluation environment.

The first part of my evaluation has demonstrated that there is a significant performance problems when using Directory Pools. I connected the test instance to my production Data Domain over NFS using the FILE devclass and I was able to get backup speeds 10x better. That doesn't really tell me though if it's a problem with the virtual environment or Directory Container Pools.

Is anyone using Directory Container Pools in a production environment, preferably with AIX? What type of performance change (assuming at least a slight drop) did you get compared to a DISK or FILE devclass?

Also, how do the deduplication rates compare to Data Domain (or similar hardware based deduplication appliances)?

 

[DazRaz]

Was your FILE devclass storage pool using deduplication? I've found that using directory container pools reduced disk IO and improved performance. No formal stats collected however. In addition, there is not need for identify processes and reclamation on the pool.

I would have a look at iostat and see which disk is struggling. I suspect your database disk(s) is not up to the job.

 

[marclant]

It will be very hard to make any conclusions about performance in a demo environment that is not sized to your workload.

Check the Blueprint v3.1 for AIX: https://www.ibm.com/developerworks/... Manager/page/IBM Spectrum Protect Blueprints
Open the PDF and go to Appendix A for performance results.

Use this table to determine if your config fits the small, medium or large blueprint:

 

[TykneeTym]

It turns out there was a problem with the underlying storage. We moved the VM to some different storage and that seemed to resolve the issue, at least to the point where the performance was consistent with Data Domain over NFS from the same system.