Hey
marc fleury wrote:
> As I said Rickard looked at it.
Yup, ran OptimizeIt during the tests.
> Rickard could you put your partial findings under the bugzilla "speed" entry
> so that we can take the discussion there and start looking at it seriously?
Basically, what I found was that the serialization of classdescriptors
took an enourmous amount of time. The only thing I could think of as a
reason for this is that currently the EJB's classloader has 4 levels:
beanloader,apploader,serverloader,systemloader. But the first one
doesn't contain any classes, but is only used for context associaton.
However, when a class is neeed, as is the case during serialization, it
will be asked since it is the first in the row, and consequently there
will be a lot of hashtable misses. And that is expensive. So, we might
want to eliminate the beanloader altogether. That *could* be it. Will
have to test more.
/Rickard
ps. the above is in BugZilla now too
--
Rickard �berg
Email: [EMAIL PROTECTED]
http://www.telkel.com
http://www.jboss.org
http://www.dreambean.com
