Bad Performance on Singleton Session EJBs ?

[rsoika]

Hi,

I am just migrating from GlassFish to WildFly. Everything works great.

But I recognized a bad performance from WildFly (8.0.0 Final) in compare to GlassFish (3.1.2) on my development machine (I am running on Linux with 64 Bit and 8GB RAM).

I investigated some time in performance analyses and I came down that it could be something with the VM memory.

For example: I usse a Singleton Session EJB. This bean uses a simple LinkedHashMap to cache some small user profile objects which I fetched from a database.

(here is the source code on GitHub)

When I call the EJB to get an object out from the cache (all objects are cached during my tests) it takes 5-20 ms running in WildFly.

When I run the same code in GlassFish the same call only takes 1ms. This is really strange.

So now I believe that something is wrong with my VM setup. I tried to configure both servers the same way.

These are my VM Options I am currently using for WildFly:

JAVA_OPTS:  -server -XX:+UseCompressedOops -Xms1024m -Xmx1024m -XX:MaxPermSize=256m -XX:NewRatio=2 -XX:PermSize=64m -XX:MaxNewSize=448m -XX:NewSize=448m -XX:SurvivorRatio=6 -XX:+UseLargePages -Djava.net.preferIPv4Stack=true -Djboss.modules.system.pkgs=org.jboss.byteman -Djava.awt.headless=true

When I start WildFly I see the following message:

OpenJDK 64-Bit Server VM warning: Failed to reserve shared memory (errno = 22).

Did anybody had made same experience or any idea what I could try to get better performance?

Thanks for any help.

===

Ralph

[wdfink]

There are several hits if you search for the error message, as example see http://www.mohancheema.net/appserver/java-hotspottm-64-bit-server-vm-warning-failed-to-reserve-shared-memory-errno-12

If you invoke the singleton, did you run multithreaded? Maybe a locking issue.

Otherwise, can you compare whether the business code in your method took this time or does it use the same time, in this case it points to something invocation related.

[kwintesencja]

or any idea what I could try to get better performance?

Hi there, i have some advices and maybe it works for you.

Singleton EJBs are transactional and blocking by default, for a cache manager i think its not the best approach, i would use the following to revert that:

@Singleton
@ConcurrencyManagement(ConcurrencyManagementType.BEAN)//means every method is non blocking so multiple threads can access it at the same time
@TransactionAttribute(TransactionAttributeType.SUPPORTS)
public class ProfileService implements Serializable{
    private ConcurrentHashMap cache;//treats concurrence

}

i hope it helps.

reference:

[1] http://www.adam-bien.com/roller/abien/entry/singleton_the_perfect_cache_facade

[rsoika]

Thanks for your comments. The @ConcurrencyManagement is interesting. But at the end it seems not to be the main issue.

I must investigate in more detail....

[rsoika]

Ok - after all there is no performance issue. After I run WildFly and GlassFIsh with same VM options and JDBC Driver they are equal.

Singleton Session EJBs are no issue.

Thanks for help.