Network load balancing

[Andy Welcomer]

I'm having problems setting up Network Load Balancing on 2 test machines
using W2K advanced server. When I configure NLB and TCP/IP, they both loose
almost 100% network connectivity. If I ping them, the machines timeout
about 80% of the time. When they do respond the time is over 1000
miliseconds. Also the virtual address
never responds. Any insight will be greatly appreciated.
Thank you in advance

Andy

 

[Matt Hickman]

I'm having problems setting up Network Load Balancing on 2 test machines
using W2K advanced server. When I configure NLB and TCP/IP, they both loose
almost 100% network connectivity. If I ping them, the machines timeout
about 80% of the time. When they do respond the time is over 1000
miliseconds. Also the virtual address
never responds. Any insight will be greatly appreciated.

Are each of your servers connected to a different switched port?

If so, you may be flooding the switch. Connect the servers in
the cluster together using a hub. and the hub to the switch.
Use unicast mode. And change the
HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Services\WLBS\Parameters\MaskSourceMac
entry to 0.

 

[Andy Welcomer]

The original setup had these machines in a hub, and I thought that may have
been the problem. I have since plugged them into a switch. Since this is
in an office area with out labeled connections I'm not sure if they are in
the same switch.

However, by changing the registry setting, the system now acting real funny.

The Virtual name is now accessible 100% of the time with fast response
times. The 2 nodes take turns going offline. One will be available, and
then a few minutes later it will go offline, and then other will be
available. At no time are they both available.

 

[Matt Hickman]

Check out the following KB articles:

http://support.microsoft.com/default.aspx?scid=kb;EN-US;193602

and

http://support.microsoft.com/?kbid=247297

 

[Andy Welcomer]

I went back to your original sugtion and plugged them back into a hub. With
the new registry setting, its working. With it working I'm now more
confused than ever.

This setup was ment to mirror and existing production setup. The production
setup is using 2 compaq ML370 servers using only one NIC and hooked into a
switch. That setup never had problems of any kind. The test that I was
working on used 2 IBM workstations with a single NIC connected to the same
type of switchs. I'm wondering, could this have been a compability issue
with the IBMs?

 

[Matt Hickman]

This setup was ment to mirror and existing production setup. The production
setup is using 2 compaq ML370 servers using only one NIC and hooked into a
switch. That setup never had problems of any kind. The test that I was
working on used 2 IBM workstations with a single NIC connected to the same
type of switchs. I'm wondering, could this have been a compability issue
with the IBMs?

It may be that the drivers for the Compaq NICs may have some code to handle
clustering gracefully that the workstation NICs do not. Servers are
more likely to be put in clusters...