The commercial Unix vendors would like us to believe that buying their high end will address these issues.

In practice, there is only so much that can be done with one box. Lavishing a machine with memory and disks will generally help but the returns eventually diminish.

Besides, the Big Iron approach still leaves us with a number of singular failure points.

The alternatives are to spread the tasks to other machines and/or deploy additional servers to handle duplicate tasks.

This leads to a different set of interesting problems. We'll discuss the approaches available to solve them.

While I'll discuss some brand names, I won't go so far as endorsing a particular commercial solution. There is a lot of new technology and potentially a lot of money involved -- I'll stay focused on evaluating the state of the art and leave dealing with the vendors up to you.