log_parallelism & recovery_parallelism

[jodie]

We are getting a new 3rd-party system and the default init parameters specify recovery_parallelism=6 and log_parallelism=6.

I've never used these parameters, and am not sure if they are necessary. From the docs, it says to increase log_parallelism if you have more than 16 processors. It doesn't really have a recommendation for recovery_parallelism.

Our system is 9.2.0.5 on AIX 5.2. We only have 4 processors.

Has anyone used these? Any recommendations?

Thanks!
Jodie

[Axr2]

I typically increase my recovery_parallelism just when setting up a standby, or during major database recovery. Setting it to a higher number (ROT of 2*CPUs) makes a significant difference in the amount of time it takes to recover a database.

[jodie]

Originally posted by Axr2
I typically increase my recovery_parallelism just when setting up a standby, or during major database recovery. Setting it to a higher number (ROT of 2*CPUs) makes a significant difference in the amount of time it takes to recover a database.

Is this true with RMAN from tape as well? If I'm restoring with one channel, does it make a difference?

Thanks for your input.

[Axr2]

recovery_parallelism has to do with instance recovery. It only comes into play *after* your archivelogs have been transferred to disk. Your TDP settings (or channel) allocation has no bearing on it. You should potentially still be able to leverage the advantages of this parameter.

[Axr2]

Just curious - but what is this tool that cares about your recovery_parallelism parameter?! I assume it has something to do with recovery.

[jodie]

Originally posted by Axr2
recovery_parallelism has to do with instance recovery. It only comes into play *after* your archivelogs have been transferred to disk. Your TDP settings (or channel) allocation has no bearing on it. You should potentially still be able to leverage the advantages of this parameter.

Duh! I'm glad it's Friday. Now that I read the description of the parameter I see that it is for crash/instance recovery. I guess it doesn't hurt to have it set.

Ever use LOG_PARALLELISM?

[Axr2]

I've never faced serious redo allocation latching issues. So, I've never had to set log_parallelism higher than the default (1).

[jodie]

Thanks for your insight.

Have a good weekend.

[tamilselvan]

LOG_PARALLELISM aids mostly in a very large OLTP system.
Orace sets the redo buffers size equivalent to LOG_PARALLELISM times LOG_BUFFERS.

Tamil

[Axr2]

Note 147471.1 has some decent info on this.