Hi,
We are currently developing a prototype using JBoss2.0-Final (hangs together
quite nicely so far - well done guys). We have 10 beans working ok and are
now evaluating performance. We run the client (servlets/resin) in separate
VMs in order to be able to cluster both client (and later on hopefully the
server).
We are following the common pattern of using session beans as facades to
entity
beans in order to reduce the number of network calls. On the client side we
are
essentially down to the following code pattern
- get the context
- get home interface of the facade session bean
- call create to get a bean
- invoke method, get data back, compose the html page
The times measure for each of the calls are around
- 100ms to get the context
- 80ms to get home
- 100ms to call create
- 80ms to invoke a method
These times are more or less overhead only. Actual time
spend in the method is on top.
We would like to speed the client turn around up and have a few questions
about what can be cached - we implemented some caching but only tested
it single user so far.
1) Getting the context once, storing it in a singleton and keep using it
seems to work well. Any downside doing this?
2) Getting home once and storing it and reusing it also seems to work
fine. Will this work with multiple clients? Ie, is the home object
always the same or does it depend on the client?
3) Caching the actual remote object for entity and statefull session
beans will obviously not work, question is with stateless session beans.
AFAIK, the server can create multiple instances to increase performance,
but are they created on invocation of "create" (in which case they
should not be cached) or are the invocation of methods dispatched to
different instances (in which case we should be able to cache the
remote object and use across different clients).
Another performance question. In order to keep beans independend, all
our entity beans are packaged into separate deployment units.
Relationships between entities are implemented using separate beans
which use client style lookups to get to the beans. Is there are
performance penalty doing this? Should we for deployment stick all
beans in one deployment unit.
Any comments anyone? Any other tips and tricks to improve performance?
Thanks for you help,
Cheers,
Robert.
--
Robert Schulz
[EMAIL PROTECTED]
--
--------------------------------------------------------------
To subscribe: [EMAIL PROTECTED]
To unsubscribe: [EMAIL PROTECTED]
List Help?: [EMAIL PROTECTED]
