Thread: Webfarm setup and NLB

I'm trying to setup a web farm using MS WFF and NLB between three servers. It's not that we need more speed, but rather the store has gone from "we have a store too..." to "Oh, this is amazing... make sure it doesn't go off line". It failed for a week and cost some serious money due to a WSI record set deleting a large block of data in error.

So we've purchased multiple machines, configured them with NLB, are using WFF 2.0 to sync up the websites. But I can't seem to find the missing pieces to make it work better. It is not as fast as the stand alone server (just a little slower), but perhaps the most notable is a very long delay getting the first page for a new web client. Something like 10 seconds before it starts to load the page. After the first page, it rips along. Our goal is 1.5 second response time, currently it averages 2.7 seconds. We plan on moving the web farm into a data center after we get the bugs worked out.

We want the ability for stateful transition if one server goes off line. Seems that there needs to be a special setup for that using a SQL server to control it, but I cannot locate information about how to set that up exactly.

We also have to deal with encrypting the config.web file on multiple machines and the certificate used to do so. In the testing environment, we are not doing that. There is an article on how to do that.

Also what is the deal with Web gardens? Would they improve performance? Why increase the number of process from 1 to a higher number? And what should that number be? If increasing the worker processes, what needs to be done to make them operate correctly in a webfarm?

Is there a good document, or set of threads to read on configuring a web farm for ADNSF? The servers are all 2008R2, NLB multicast with multiple nics, ADNSF is 8.0.1.2 I believe the NLB set is working okay, not sure how to tell if it isn't. I can add or drop servers from the NLB and they seem to take load just fine once the connection is started.

Any thoughts on this topic?

Surely none of this would have helped you in your WSI deleting data scenario.

The webfarm + NLB will give you redundancy on web server level, but your mssql
server will still be a single point of failure (if the mssql goes down the web farm goes down). You have to also maintain a state server with separate mssql database for
asp.net state sync between web servers.

From our experience webfarm+nlb is great for high-traffic sites where you have to distribute the traffic between multiple web servers.

For fail-over we use Windows Server 2008R2 with Hyper-V and iSCSI storage. Each site is running in a hyper-v VM (with portable VLAN IPs), in case of hardware server
failure the VM automatically starts on another Hyper-V Host. The downtime with this
setup is usually less than 5min.

Well, no none of the web farm stuff would have mattered to the WSI deleting. But it did emphasize the need for the site to be up 100% of the time. The site has gone from a weekend fun project to a serious source of revenue in the last 7 years, and when it was more or less off line for a week it drove the point home. This platform was the only serious contender for the web software since we have over 600,000 SKUs.

We will duplicate the SQL server in a cluster eventually, but at the moment that is running on a very expensive box.

I'm specifically looking for details on what needs to be done to support the web state server, and information on how the web garden would be deployed and whether it would improve performance.