JBossDeveloper

Agreed. Enabling TxStats in a performance run is like enabling debugging!

There's already a JIRA around synchronised methods. But if you've got performance results that illustrate areas of concern (with the right configuration options enabled ) then attach them to the root performance JIRA that I mentioned.

The Xid wrapper PR is 404 for me too.

What's the before and after view of the class look like? No, I don't want to spin up Eclipse

OK hold on ... your original comment mentioned skipping a step around the bqual but now you've deleted the class entirely in your codebase. I must be missing something. Please explain.

Jesper Pedersen wrote:

 

Scaling

 

However, I'm currently blocked on what look to be a main lock somewhere, as the transactions per seconds drops as soon as there is a load on XAResource.start() and XAResource.commit() - e.g. the sleep is != 0. It doesn't seem to matter what the underlying storage implementation is, as both the memory store, and the file store gives the same numbers. But maybe my configuration needs to be adjusted,

 

Mem store - https://github.com/ironjacamar/ironjacamar/blob/1.2/embedded/src/main/resources/transaction.xml

File store - https://github.com/ironjacamar/ironjacamar/blob/1.2/core/src/test/resources/file-transaction.xml

 

The attached graph shows this -- the "missing" curves are on top of each other. Any ideas on this would be great.

 

As you say, it looks like you aren't successfully enabling the volatile store correctly hence the results are the same which is certainly not what I would expect for 2PC transactions. Here is an example of configuration to enable the JDBC object store as an example: narayana/ArjunaJTA/jta/src/test/resources/h2jbossts-properties.xml at master · jbosstm/narayana · GitHub

You can find out if you've enabled the VolatileStore by calling the appropriate method of StoreManager.

Hi Jesper,

I see we are up to 5 PRs now for the enhancements. I think it would definitely be useful to have a thread each for these and to discuss each one separately rather than on this main thread as following it is getting confusing.It would also be really great if you could give us an idea of the relative performance benefit for each enhancement that you have identified. I personally thing that although small improvements are still improvements we should get a feel for which improvements are going to provide the greatest wins and focus on those.

Tom

EDIT: Just to clarify too, when I say a different thread I also mean it would be useful to capture each as separate Jiras too so we can audit each commit - thanks

Very strange. We have some raw performance tests in ArjunaCore/ArjunaJTA so I may take a look at these tonight if I have time. You should definitely be seeing a significant performance difference between Volatile and Log.