I'm currently running a WAS plant with the following topology:

F5 Load Balancing > IIS Web Servers (hosting plugin, round robin) > WAS ND (RHEL)

I was asked to POC removing the IIS layer in between.  This would allow F5's to handle session affinity, and url redirection. 

We have Siteminder as our policy server which is configured at the IIS layer.  Therefore, removing IIS is out of the options since F5 cannot perform the Siteminder auth work.

Someone from the F5 team advised to have 1 physical web server (instance of IIS) fronting only 1 JVM.  This would mean in a production environment, we'd have 6 different web servers assigned to their own unique JVM > 6 web talking to 6 jvm's.  Not sure how horizontal clustering would be impacted with this, also would multiple plugins be required for this implementation, other impact?

I'm looking for thoughts on this type of deployment.

Thanks in advance for your thoughtful time,
Webster

Good morning Webster.

I'm not sure why someone would recommend a one-to-one relationship of web servers to JVMs behind an F5...

I happen to run an environment similar to yours where WebSphere is front-ended by IIS web servers and a pair of F5 load balanaces.

In my configuration, I have numerous WebSphere installations spanning quite a few hosts.  Each application in WebSphere uses a web site in IIS to route traffic to the application/ websphere cluster (I'm running ND.)  Ahead of that is the F5 that hosts the VIP (virtual IP) that the DNS entry resolves to.  The F5 then forwards the request to the IIS webserver defined in it's "Default Pool" (F5 terminology) and uses the values defined we defined for the the F5 "Persistence".

The web servers we run are IIS, as previously mentioned, but they are shared. I perform a custom installation of the WebSphere PlugIn on IIS (I don't use the installers) to do what I want.  It's essentially creating a new dedicated website in IIS, then isntalling a new isntance of the PlugIn to run for that website.  I ultimately create the web site in IIS and then map my applications to include these resources and generate the plugin.  I then push the plugin to the IIS website.

This approach lets me load balance (via F5) across a small cluster of IIS web servers (which are redundandt) and route traffic to the correct WebSphere instance.  The benefits include web site redundancy, hardward loadbalancing via F5, individual websites and plugins, etc.

My configuration my be a bit different than what you aim for as session persistance is at the JVM for the applications I'm hosting.

Hopefully this proves useful!

Erik 

One  question, your WAS configuration is  using cluster, because  you can build  something like that:

                                                         siteminder
                                                             |
Virtual Server  --> F5 --> was_pool --> IIS1(active) -->  WAS ND cluster
(listen port80)                               |--->IIS2(passive) --> WAS ND cluster
                                                             |
                                                          siteminder 


About custom plugin configuration for  have  more that one webpshere jvm on each IIS,  I do something like that  using IHS  on linux, just be careful  when apply changes in plugin  configuration files.

If you dont have  cluster array, maybe you can put your IIS harden in front of bigip and connect  your VM's to the pool members(but you need think in performance and  security first). 

hopefully you find it useful
Dario Garcia