Oracle Observations

December 4, 2008

The RAC roundtable

Filed under: Uncategorized — bigdaveroberts @ 9:55 pm

Please note as a non-expert, I don’t guarantee my ability to accurately summarise everything!

The experts were Julian Dyke, Joel Goodman, + 2 others.

The first question concerned connection pooling, and the unbalanced load experienced wile attempting to load balance.

Several issues were discussed while closing in on a conclusion.

Joel explained the new FAN events covering service goodness that have been introduced in 11gR2. However these new asynchronous notifications are only used if the JDBC ICC client is used or if the middle tier was aware of these notifications.

As the person asking the question appeared to be setting sensible min and max handlers, it was concluded that the issue was that load balancing was occurring at connect time. I.e. Differing numbers of connections are established at connect time to the RAC servers based on current load, whereas the goal should have been uniform connection allocation initially, and then load balancing needs to be established on the basis of allocation from the existing pool on the basis of current load. Essentially the goal wasn’t to increase the number of connections to notes that were being underutilised, but to rather use more frequently the connections that already existed to the nodes that were under utilised.

Key to achieving this runtime load balancing was utilisation of services to handle listener connections.

There was a brief associated discussion as to which connect time algorithm should be used.

The conclusion was that clbgul (?)=long should be set, although doubt was expressed that this actually related to the original question.

A curious question was then asked about RAC in a virtual IBM server environment, Doubt was expressed that RAC was actually supported in that environment, but that doubt seemed to ebb away.

The question was, should the virtual RAC environment be given another node, more CPUs in existing nodes or an additional CPUs in the existing nodes.

The simple answer was
1) Better CPUs
2) More CPUs
3) More nodes.

However then came the caveats.
1) The amount of batch work may influence the decision.
2) While faster CPUs is traditionally considered the solution that involves less risk, there is the possibility that the increased throughput might increase load on the interconnect, which is probably the biggest performance inhibitor. Ergo there is no simple answer!
3) As the environment was virtual, then it was probably worthwhile just trying it and seeing what the result was!
4) It was pointed out that if the reason for going RAC was HA rather than performance, than all the above answers were wrong! As the main issue would be by going to a 3 server RAC cluster, the user benefited from the fact that a node failure would result in the loss of 1/3rd of the available capacity, rather than ½ that would currently be experienced.

There was a brief discussion of interdependence of clusterware, asm and rdbms versions, and from what point rolling upgrades were available.

Clusterware has always supported rolling upgrades.
ASM supports rolling upgrades from 11g.

Finally the version of Clusterware should be >= the version of ASM >= the version of RDBMS.

Next the question of RAC on GPFS vs. ASM on AIX was asked.

Many questions and observations about technology and politics of file system choice were tackled before attendees attested that RAC worked well on both!

Before that answer was established, the fact that ASM didn’t suit SAP was observed, because SAP requires access to a native file system.

ASMfs would e introduced with ASM 11.2 and would expose the ASM as a normal file system.

An advantage of ASM was that it opened, held and cached data file file handles, which was an optimisation that wouldn’t be possible with other file systems.

Utilisation of the white papers produced by the MAA group was recommended.

What are the effects of internal redundancy?
Memory, IO and CPU! In that order!

The issue of block versioning in RAC was covered.

Essentially multiple versions of blocks can exist on different nodes. When writing, this isn’t a consideration. The latest version of the block needs to be either read from disk, or passed from another node in the cluster causing traffic. However for reading, out of date duplicate blocks on the current node can be utilised for logical reads!

Again, as with the security round table, only the surface seemed to be scratched!


1 Comment »

  1. Interesting stuff indeed! =) Thanks for posting.

    The whole “rolling-upgrade” issue is extremely misleading, IMO. After having done a couple upgrades from to, I found out the hard way what the PR really means to those of us who do all the heavy lifting. The original documentation suggested that one could roll from a “supported” version of 10gR2 (ie, to 11g. This is only true for CRS, not ASM, even though the documentation states otherwise. Apparently, the rolling portion of ASM is supposed to be available in 11gR2, and the last time I checked, 11gR2 was not out yet. And even though a rolling CRS upgrade is possible, the documentation is at times very vague or just plain inaccurate. I have had a good exchange with the documentation team and they were working hard to correct those issues.

    It will be very interesting to see how Oracle’s ASM continues to evolve, and how the whole market will react in the next couple years. At our workplace, I am very curious to see which job roles finally get “lucky” enough to add ASM to their job description. =) (I for one still have a lot to learn in that space)

    Comment by Charles Schultz — February 6, 2009 @ 2:30 am | Reply

RSS feed for comments on this post. TrackBack URI

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Blog at

%d bloggers like this: