JBoss Clustering + Windows NLB (Network Load Balancing) = Ba

[evk]

Hi,

I recently tried setting up 3 machines in a JBoss cluster that were ALSO in a Windows Network Load Balancing (NLB) cluster. I attempted this because I figured NLB would give me a virtual IP address that I can use to load balance HTTP requests to members of the JBoss cluster (serving out functionality via stateless session beans and web services)

This seems to have failed miserably; Jboss locks up on startup and exhibits other strange behavior until I turn off NLB.

Is this Not A Good Idea or otherwise Officially Not Supported?

If all I want is HTTP load balancing for fault tolerance (high availability), is a hardware load balancer probably my best option? I tried mod_jk and it seems ok, but now the machine running Apache represents the single point of failure.

[brian.stansberry]

We haven't had a lot of community input re usage of NLB, and we test with it or use it internal within the JBoss division, so that makes it somewhat a Not Good Idea since it will be harder to get community support. I don't have any real idea what your problem could be, only than I found a probably irrelevant post via google:

http://www.nabble.com/TCP-startup-problems-t563478.html

A quick glance at NLB docs make it look like its support for session affinity is not so hot. Also, it uses client addresses to pin users to a server. If your clients are sitting behind corporate firewalls, you could have a lot of users who appear to be using the same address. They'll all get pinned to the same server.

Re: mod_jk vs. a hardware load balancer, the hardware load balancer is also a potential single point of failure (unless you get two, and you can always get 2 apache boxes as well).