Parallelising App server architecture

Question

Hi:
I m fairly new to the J2EE technologies. ( thought I ve been reading
about them for the last couple of months or so. )
I m trying to understand how an organisation can scale a J2EE
application, when its load increases. Essentially expanding
breadthwise ( more parallelism ).
Basically the most distributed architecture is this
1. Weserver runs in machine A, it forwards the requests from the
client web browser to Servlet engines running on Machine B. Machine B
delegates the requests to EJB containers running in Machine C. From
here it can go to machine D where database engine is running. Is it
possible to expand this any further.
2. Also in teh above scenario, for web servers ( running in A ) to be
able to forward the requests to Servlet engines ( running in B ), a
small plugin from teh Servlet engine vendor needs to be installed. (
Pl correct me if there is a standards beased/portable request
forwarding mechanism/protocol from the webserver to the Servlet engine
) ( I heard a protocol term with respect to tomcat - can t remember it
now - but it was specific to tomcat IIRC )
3. Once in Servlet the only was I can send a request to EJB running on
a container on machine C, is by using the ejb home object through a
JNDI lookup. Or in other words, there is no automatic mechanism like
teh web server-servlet engine bridge, to forward the requests from a
servlet to its EJB.
4. If 3. above is true, the only way to load balance between multiple
EJB containers running in multiple machines ( C1, C2, C3... ) is by
hosting different home object references on the JNDI tree, and to
fetch the references in some round robin fashion.
5. By using round robin DNS or web server switches ( CISCO or other
hardware that connects to different machines for teh same IP address
in a time shared fashion ), in general it is possible to load share
between multiple web servers, even if client requests belong to
sessions, that might require all the requests belonging to teh same
session go to the same web server.
6. Also though there might be vendor specific mechanisms, how it is
acheived, it is in general for a given web server to load balance
between multiple Servlet engines ( B1, B2, B3 ... )
My ultimate goal is to see if it is possible to have an M:N:O type
configuration ( M web servers, N Servlet engines, O Ejb container
hosting machines )
Any answers even parts of it will be very helpful.
Thanks
--sony

Answer 1

Please don't cross-post.
4. If 3. above is true, the only way to load balance between multiple
EJB containers running in multiple machines ( C1, C2, C3... ) is by
hosting different home object references on the JNDI tree, and to
fetch the references in some round robin fashion.No you don't have to load balance yourself.
When clustering is configured on an EJB, the home object you get from one
server, and it knows what are the other servers. It will load balance the
requests for you. It's called "cluster-aware stubs". I call it "smart
stubs".
===
My personal opinion regarding clustering; don't try to make a complicated
configuration. You need load-balancing? Use up-front (Web) load balancing
and then stick to a single server in the cluster, containing everything.
This will make your everything easier: configuration, deployment, upgrades,
troubleshooting, etc. Access your EJBs locally for more speed.
==
Emmanuel