All of the documentation states that load balancing should be a trivial matter since the architecture is effectively stateless; however, I am not seeing this in my test environment.  Here is my setup:

 

Server #1 - BOS running under ServerConsole.exe (Windows Server 2003 Sp2)

Server #2 - BOS running under ServerConsole.exe (Windows server 2003 Sp2)

Client machine - running Winform app (Windows XP)

Database server running SQL 2005 Sp2 (Windows Server 2003 Sp2)

 

Both of these servers have MS Network Load Balancing configured with a Virtual IP to balance port 9009 with Affinity set to None.  The Servers are listening on Port 9009 using the http virtual IP address of the NLB cluster.

 

With Affinity set to None, each "request" from the client to the server should be balanced so as I use the application I should be seeing connections on each machine (using MS Port Query to trace connections).  I am not seeing this.  Additionally, if I close the application on the client and re-open I am always connecting to the 1st server to have received the initial request from my client.  Furthermore, if I close down ServerConsole.exe on Server #1 (the one receiving all of the connections), the application exceptions.  I see the same behavior using Single for the Affinity so if I shut down the serverconsole.exe on Server 1, the client should be balacing to server 2...no luck.

 

If anyone has any suggestions I would appreciate it.

 

Thanks

 

 

 

This would sem to be a problem with the configuration of your Load Balancer.  There is nothing that DevForce could be doing that could explain this behavior.

 

David

 

I'm actually using Microsoft Network Load Balancing (software load balancing built into the OS).  I actually have multiple clients balancing now but I'm having issues with clients being able to failover when one of the load balanced BOS server becomes unavailable.  I've configured the load balancing with multicast support since I'm only using a single NIC.  I am continuing to investigate whether the NLB software is correctly communicating (and removing that node from load balancing) with each node when a load balanced port (9009 in this case) becomes unvailable.

 

I appreciate you looking into this.

 

I have however been unable to correctly load balance multple requests from a single client when load balancing Affinity (or Sticky) is set to None.  Whether this is possible with the persistence service I'm not sure. 

 

Thanks again.

 

Rich