Video Screencast Help

How to get better backup throughput for RMAN backups to NetBackup?

Created: 06 Nov 2012 • Updated: 22 Dec 2012 | 7 comments
This issue has been solved. See solution.

We are using the RMAN NBU agent to backup Oracle DBs. Data servers are RHEL, as are the media servers, using NB 7.1 to backup to Data Domain via OST. We have capped the number of streams for the RMAN backups at 8, however, during some benchmarking, it appears the DB server can handle more streams, and hence get better backup thoughput. The problem is, we need to cap the streams since we have a limited number on the DD side to accommodate all of our backup jobs. So, is there anything I can tune (either on the RMAN side, or NB side) to send more data over the streams?

Our disk buffer size is set to 262144, and the number of buffers is 64  if that matters.

Comments 7 CommentsJump to latest comment

gc_bus's picture

I suspect the problem might be on your Data Domain side of things. We use Puredisk for back-end storage with 50 concurrent jobs allowed per pool. We have a 20GBPS network connection from the clients to the backup media servers and then 2GBPS to each of the content routers in each Puredisk pool.  Our buffer size and number of buffers are the same as yours.

With this setup we can easily backup Oracle databases at 1GBPS per client if the client is performant enough.

LT2000's picture

In reading through the oracle docs, it seems that increasing the FILESPERSET would increase the amount of data per channel. But DD best practice for Oracle is to keep that setting to 1.

gc_bus : I don't think our network is the bottleneck since we can get better backup throughput by increasing the number of channels. I am trying to ascertain whether we can increase the amount of data transmitted per channel. It appears I can at the expense of dedupe efficiency.

Nicolai's picture

I don't think the Data Domain is the isssue. They are very fast - Unless the network cables have some sort of fault.

Try the GEN_DATA directive to verify you have the required bandwith.

With 4 media servers I get 800MB/sec on a DD890.

Assumption is the mother of all mess ups.

If this post answered your'e qustion -  Please mark as a soloution.

LT2000's picture

Nicolai thanks for that tidbit of learn something new everyday!

I'm thinking about using that directive to test throughput without needing the DBA.

Nicolai's picture

And for that GEN_DATA is a great tool. Basically you can later lean and tell exactly what the limits you're Netbackup installation.

I would suggest fist running GEN_DATA from the media servers and see if you are getting consistent throuput values. Then run GEN_DATA from the client. This will tell you what the limits are for the entire transfer path.

Assumption is the mother of all mess ups.

If this post answered your'e qustion -  Please mark as a soloution.