My two largest customers suddenly both decided they wanted to stop playing around with a test MOSS server, and put SharePoint into production. So all week I've been building and configuring small SharePoint farms. I ran into a real headache today, which I'll discuss below.
First, here are the two deployments:
SharePoint Farm with No Load Balancing:
This solution means you have a single web front-end server, but multiple servers in the farm, mostly for distribution of load. This customer wanted a dedicated indexing server, with the index replicated over to a share on the WFE server.
Figure 1: Generic SharePoint Farm without Network Load Balancing
SharePoint Farm with Load Balancing:
This customer decided to go with load balancing, which means you have multiple WFE servers all serving the same content. In this case we did not set up any dedicated servers; the indexing and searching is distributed across both MOSS servers.
Figure 2: Generic SharePoint Farm with Network Load Balancing
Let's face it: Information Technology is a large enough discipline that it's easy to put yourself in a situation where you go from respected expert to bumbling moron in the blink of an eye. It's bound to happen, since SharePoint crosses so many technologies. Today, it was something as seemingly simple as setting up a two-node NLB cluster.
I'm not going to cover the specifics of setting up NLB, because that is covered elsewhere in lavish detail.
In NLB, there are two modes for traffic: Unicast and Multicast. Unicast basically means the network card is dedicated to NLB traffic, and Multicast means you can use it for both NLB and regular traffic. In the second figure above, the two MOSS servers each had one network card, so my plan was to use Multicast. Everything converged beautifully and the clouds parted and the angels sang. I got MOSS installed and joined to the farm. I was able to browse to the "Portal" site hosted on the clustered IP, from several different servers.
However, I couldn't reach the site from my laptop. Or even ping the clustered IP. Nobody else could, either. Some of you are already nodding your heads. Bonus points to you.
We found out that all the machines on the same subnet could connect to the clustered IP just fine, but nobody else could. Since this is an enterprise with numerous subnets, this presented a problem.
It turns out that due to the way Multicast works, some gateways do not route its traffic properly. Older routers can't handle Multicast at all. Newer ones can, but this functionality is often turned off. So in the end, we added another NIC to each MOSS server (luckily these are virtual machines, so it was easy), switched to Unicast, and the problem was resolved.
So here is a rule of thumb: Multicast is fine if you have a single subnet. But use Unicast if you have more than one subnet (or a WAN).
As soon as we switched over and reconverged, everything worked wonderfully again. The clouds didn't part this time, but I still felt pretty good about it.
0 comments:
Post a Comment