Tonight’s class talked about clustering — the combining of two or more physical computers into one network node. The advantage of doing this is that it lets “one” server take the brunt of any activity you wnat to throw its way. In this context, we used it to set up a “single” mail server that was actually made up of two servers that shared the responsibility of the packets that showed up for processing at the node.
We talked about two distinct types of clustering. The first, the Microsoft Clustering Service, was described as the “big daddy” of clustering. In simple terms, it’s application aware, and determines not only which server is the least busy, but it determines which physical server would be the most appropriate to handle any given single packet. Using the MS Cluster Service, you can have anywhere between two and eight nodes in your cluster, and you have to have specialized hardware designed specifically for clustering.
The “diet” version of clustering we discussed is called Network Load Balancing. With NLB, you are allowed between 2 and 32 nodes and you can use any Windows 2003 server without specialized hardware. The downside is how it handles its load balancing: it ignores what each server is “good” at and routes incoming packets to the server that’s the least busy. Technically, it does balance the load, but sometimes that balancing leaves something to be desired in terms of efficiency.
After our cluster — ahem — we moved on to back-end and front-end servers. In Exchange, every server is considered a back-end server by default. A front end server is one that is exposed to the Internet — so, as such, we need to make sure we take certain security precautions on our front-end servers. Specifically, we need to not have any mailboxes stored on the front-end servers — we store those on the back end servers and let the front-end servers divert traffic to them. The back-end servers sit safely behind our corporate firewall, where life is cozy and warm.
Homework:
- Chapter Four: Review Questions