Tiered Distribution

As I was looking for information on MS Clustering Services ( MSCS ) (Failover Clustering ( WSFC ) in Windows Server 2008) and how it related to the overall picture, I kept digging deeper and deeper into the ‘why’ you would do it this way (cluster servers). I found out the answer lies in Tiered Distribution. Before I could dive into failover clustering, I had to first remember back to my college days about tiered architectures. I did well on the exams, but spitting out answers to a question does not prove an understanding of the subject, the ‘why’ if you will. So in order to tackle my MSCS / WSFC questions as they relate to SQL Server 2005, I had to first make sure I knew ‘why’ best practice infrastructures involve tiers. Here’s what I found out:

Single Tier – Mainframe

Back in the day when IBM ruled supreme on anything computer, there was basically only one tier: the mainframe. The mainframe handled everything: applications, data, etc. The two main problems with a mainframe is scalability and expense. Mainframes have a physical limit on memory, processing power and disk space. Once you’ve reached the limit, the next expansion will involve buying another mainframe, which brings up the expense issue.

Two-Tiered – Clients and Databases

The logical progression from the mainframe is to increase scalability by pushing application processing to the client. In this way you can add new machines at small expense and also tweak each user’s machine to their specific needs. Users no longer have to share processing power with everyone else. Drawbacks include the management of these clients: you better hire a ton of desktop guys. Also, you can’t let anyone outside of the company use your system.

Three-Tiered – Clients, Applications, and Databases

The next evolution in tiered distribution is a compromise between the mainframe ease of having few machines and the scalability of a two-tier distribution. As you can see here, the application work has been brought to a middle tier called the application tier. Users usually make requests of these applications through a thin client, and the web and application servers do the work. This is what stumped me in school, because the difference here was not explained well. You wouldn’t move applications like Word or Excel to the app server because they are single user applications, but you would move applications like PeopleSoft, Remedy or Documentum because these are collaborative effort applications. One of the biggest problems with the three-tier distribution is the application tier is now exposed to external customers, so they can connect to the web servers. Not a great idea.

Four-Tiered – Clients, Web Servers, Applications and Databases

Ah this looks much better. With the addition of the web server tier, we can now put our web server in the DMZ or perimeter network without exposing our application servers to the outside world. Additionally we can give the two tiers much different set ups. We would give the web server tier more network sockets and I/O while we give the application servers more processing power, which the web servers don’t really need.

Conclusion

Hopefully this makes it very clear as to why tiered distribution is the current norm in a typical infrastructure. It really made it easier for me to understand why you would cluster SQL Server as well as how it actually works. For more in depth information on this subject, check out this Microsoft best practices page on Tiered Distribution.