RMI LoadBalancing and failover without an application server.

Question

Hi All,
I have just one RMI object doing mathematical
calculations.I want to provide
1. Load Balancing and
2. Failover
for this rmi object.Few Application
servers(like weblogic) provide a wrapper over sun's
rmi to produce replica aware rmi stubs but they force
to use vendor specific (like weblogic rmi-registry)
rmi-registry, which runs within the app server.
My problem is :
I dont want to run the whole app server to just run
one rmi object ?
Is there a way using which I can provide load balancing to my single rmi object ?
Thanks in advance for the help !!
Regards
Rais

Answer 1

To provide load balancing the best way that I know of would be to create a front server that handles distribution of work between several back servers. This would be how they do it in app servert load balancing to distribute the load amonst several app servers like BEA Weblogic, Microsoft IIS and others. So create a "dummy" from RMI server that:
1. Queries the servers that it knows about and finds out how how many request it can handle and how many request it is currently handling (they would keep track internally).
2. Use an algorithm to determine which server to send the request to
3. Pass the request through.
What a request is would probably be best designed as an interface like IRequest so regardless of the addition of IRequest i the system the RMI back end servers are sure to be able to process it. How ever ther will always be a maximum number of request that can be handled that can only be changed by adding more resources (RMI server machines).
RMI request ---> RMIServer ---> RMIServer 1 on machine 1
---> RMIServer 2 on machine 2
---> RMIServer 2 on machine 2
You would most likely want to have a base interface that defines a Server in your system like IServer because the front RMI server from above is really nothing more that a regular instance of your server implementation except configured to be a load balancer.
Possible configuration file for the servers(server.conf):
Config for the front load balancing server would look like this:
loadBalancer=1 #Tells the server to use a LoadBalancerRequestHandlerImpl from startup
balanceServers=http://machine1:port/server1,http://machine2:port/server2,http://machine3:port/server3
All Other servers like this
loadBalancer=0 #Tells the server to use a regualar RequestHandlerImpl from startup
balanceServers=
The about RequestHandlerImpl,LoadBalancerRequestHandlerImpl are just some ficticious implementation of you Business Process stuff like login,add user in User database app. etc.
FAILOVER
By failover I assume that you are refering to if an app fails then who handles it's request that it currently has. Well you can do this qite simple by making the back end servers (request servers) knowledgable about the fron server(s)(load balancers). Then when ever (assuming a nice server) will check it's config for who it's load balancer(s) are and hand over the sessions.
Well since this is RMI and assumig that you are not doing login stuff the above is not practical. The load balance servers(s) would may be in a back groud thread(low priority) try to comm with it's servers from the balanceServers property (from the earlier config file, now in memory in some data structure) and if comm failes with a particular server then remove it from the list. THen it will no longer be eligible to recieve request.
I ama bit long winded in my explaination but I hope this helps.
I hope that