5.6 Cluster failover architecture

Many people think that all the excitement about clustering is to prevent system down time due to some type of catastrophic hardware failure. In the old days, that would not have been a bad assumption. When we got into the computer business over 20 years ago, a VAX computer was built up from 30 or more large printed circuit boards, each much larger than a PC's motherboard today. After performing some extensive mathematical calculations, we can tell you with confidence that at some point that system would crash and that the only mathematical law that mattered was Murphy's law. With such a large parts count and all the wires and connections necessary to interconnect all those logic cards, it was no surprise when one of those modules failed. The only question back then was when would it fail? Today's servers are built with a single motherboard and maybe 4 to 10 adapter cards. The CPU is now a single piece of silicon as opposed to hundreds of discrete components. Because of these advances in technology, the reliability of today's computers is much greater than when we got started. Now we are not going to try to convince you that computers don't crash today, but surveys conducted recently point to other causes of system downtime these days.

Cluster failover is probably going to be the most difficult issue to deal with for a system administrator because you will have so many options available to you. You have...