MS NLB App Center 2000 on VI3

[Rob_Bohmann1]

I migrated an application we use for self service password resets from our 2.5 hosts to our 3.0 hosts. Initially we had some problems where the application was not working. The cluster was not getting established and the app was not working. The resolution was to disable the cluster nics and then re-enable them. This application uses unicast NLB.

Today the web site (cluster ip) was down again. I suggested the app owner disable and re-enable the nics again cause it worked last time. It did not resolve it. He powered down one of the cluster (secondary) for about 5 minutes and powered it back up and the application started working again.

So now he reports that he can ping the cluster ip address between the hosts, but when he tries to ping host 1 name to host2 name or vice versa he gets no response.

When he powered on the secondary host I see an entry in the vmkernel log near the same time that states " Starting world vmware-vmx with flags 44 ".

The application is up and running for now. It has always been a tempermental ..... server/application.

In searching thru the knowledgebase I found articles about NLB but nothing specific to VI3 or if anything was different between 2.5 and 3.0.

So my question and my asking for suggestions or advice is "Has anyone migrated from 2.5 to 3.0, an NLB based application cluster and did you have any problems?" If so can you describe and advise please? Or if you have other thoughts and suggestions that would also be appreciated. My experience with NLB is nano-sized.

[masaki]

I've seen this on ESX 2.5 servers.

We had an NLB cluster and everytime we vmotioned the vm they loose connection.

If we remove the vmnics and then recreate them All was ok.

I think NLB is unsupported.

[bister]

We have this issue too. We now run all NLB nodes on the same ESX host. The problem is that all NLB nodes share the same virtual MAC and moving (Vmotioning) this virtual MAC over routers causes the system to fail. This is not a ESX problem, it's a general network issue and expected. VMotioning all NLB nodes together from one host to another should work.

[Rob_Bohmann1]

Thanks for the replies. Yes we are aware of the results of vmotioning the servers.

So Bister, if you vmotion both servers in an NLB cluster at the same time from esxhost1 to esxhost2, this does not break the cluster? Or am I misunderstanding you?

Any ideas on why we would not be able to ping the other clusters host name from one NLB host to the other?

[bister]

This is the information our guy who is managing the NLB cluster (it's a Citrix Secure Gateway server cluster) gave me. So the answer is: Yes.

I have no further idea.... sorry.

[Rob_Bohmann1]

thanks i will have to test this to see if this works for our application.