10-07-04 10:49 PM
Brent Friedman wrote:
> I am trying to set up a test environment to test seamless failover for a
> Java application. I have looked over everything on WebSphere
> clustering that I can find, but I still have a few questions.
>
> Hopefully, someone here is a clustering/failover guru, and can answer
> these questions in their sleep. ;)
>
> 1. I have looked at the HTTP plug-in, Network Dispatcher, and some
> other methods of clustering WAS. It *seems* that the HTTP plug-in
> provides session failover, but the user would have to re-authenticate to
> get back to their session. Is this true?
Not at all, no. If you wish to continue using the same session on a new
application server in a failover situation you would need to configure
some form of session persistence (either to a database or using the
publish/subscribe model). There would be no re-authentication required.
>
> 2. Would Network Dispatcher provide seamless failover? (Specifically,
> user A logs into an intranet site. They are authenticated, and routed
> to WAS node 1. In the process of going through screens of the web app,
> WAS node 1 loses connectivity. Will Network Dispatcher sense that Node
> 1 is down, and automatically re-route the user to node 2? Or, does the
> user have to click back button in browser, reload page, and/or
> re-authenticate?)
Don't use Network Dispatcher for this, the HTTP plug-in is the correct
solution. Use Network Dispatcher to spread load across your HTTP servers.
>
> 3. Would a Tivoli Webseal junction assist with re-authentication in
> case of failure of a particular WAS instance?
Not required
>
> 4. What is the "typical" WebSphere component mix to enable seamless
> failover?
Some form of IP sprayer such as Network Dispatcher, obviously with a hot
standby. This sprays to at least two HTTP servers. These send work using
the HTTP plugin to at least two appserver nodes. Recommend that you use
session affinity. You may also wish to use session persistence depending
on the need to recover in-flight sessions during failover.
Also think about dMgr failover - this is a potential single point of
failover, you need some solution for this.
[ Post a follow-up to this message ]
|