cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

AppDynamics and batch processing (Springbatch) / performance degradation

Stijn.Schepers
New Poster

Hi Community,

Has any-one experienced a serious performance degradation while monitoring batch processing with AppDynamics? The application we are monitoring is a java app (spring batch) runnning in WebSphere. The batches are being processed with multiple parrallel threads (>10). We see that with AppD monitoring the duration of the batch slows down with a factor 2. We are using appd 4.4.1 and are running in production mode (not in Development Level Monitoring).

This must be a config issue? Anyone had similar issues and how to solve it?

Thanks,
Stijn

 

7 REPLIES 7

Peter.Holditch
Moderator
Moderator

Stjin,

That is definitely not expected!

 

How is the heap sizing?  If the JVM was operating close to heap capacity before you introduced the agent, maybe the additional agent memory footprint is now causing frequent Garbage collection?

 

If that's not the case, then we need to figure out what is consuming the time.  Can post some agent logs, either here or in a support ticket?

 

Warm regards,
Peter



Found something helpful? Click the Accept as Solution button to help others find answers faster.
Liked something? Click the Thumbs Up button.

Thanks. First time I've noticed a real difference but I have been mostly involved with online transactional processing (not batch driven processing). We do see a significant overhead of CPU utilisation from AppDynamics (between 5-10%). This drives CPU Busy Time to 100% resulting in batches running way slower. But CPU Busy could be also a symptom of something else. Opening a support ticket and will feed back the result to the community. Hopeful this is just an easy to solve config problem .

Thanks

Stijn

We have done more analysis and we can see that the overhead of AppD for this application is 10% CPU Time. The JAVA App runs in WebShere on a Virtual Machine on an ESX server. Quite classical setup. 

 

It seems that our selection of Service Endpoints had a dramatic impact on the performance of the application. So be careful when setting these in your instrumentation.... !

 

This is a very surprising finding to me.

 

What was your SEP configuration before and after?



Found something helpful? Click the Accept as Solution button to help others find answers faster.
Liked something? Click the Thumbs Up button.

We had springbatch included. By disabling the spingbatch option, the duration of a springbatch run was faster and the cpu utilisation decreased. We had 10% overhead with this option which is not acceptable. Suprised as well as the documentation talks about a very slighy difference in CPU utilisation by defining end-points.

Configuration attached