In apache 2, the MaxClients directive in your apache configuration will determine the maximum number of server processes allowed to start. As you probably guessed, it's a very important directive. If you reach the limit, you'll get a very friendly message in your apache error log saying:

[Thu Oct 18 06:22:19 2007] [error] server reached MaxClients setting, consider raising the MaxClients setting

Well, the error isn't so bad but what the user gets is anything but friendly. Basically, the browser waits, and waits, and waits until a process frees up. This can be very quick or it can take a long time depending on what those other processes are doing. The end result is that your server is unavailable (a.k.a. down) until resources free up. Not a good scenario.

The default on Debian for MaxClients is 150 last I checked, which is actually a pretty good starting point. The temptation is to change this to a very high number, for example 1000. The problem with a high number, however, is that it assumes that hardware resources can support that high number. In other words, apache will continue to start new processes until it reaches that number. For a hardware-constrained apache, this can result in disaster. You may end up with a system unable to keep up with memory demands where only a reboot will restore things. This is known as thrashing. When a system thrashes, it is so busy swapping memory to and from the hard disk while processes are continuing to queue up that it has no time to actually get anything done. In some cases, a really fast shutdown of apache will save the day. But if the problem has gone on for long enough, you will be lucky to get a shell prompt.

You can estimate the number of MaxClients your system will support by taking the amount of memory and dividing it by what you expect each apache process will need. If you have 8GB of memory and you believe apache will need 20MB per process, 8192/20 = 409.6. You need to consider the fact that other programs will need memory too. If you have Apache and MySQL running on the same box, do you think MySQL processes will stay the same when Apache gets busy? More likely, they will both eat up resources during peak hours. If you leave MySQL with half the available memory, raising MaxClients to 200 should be a safe decision. You're better off erring on the side of caution. You can always raise the number later.

To keep things working smoothly, you probably want to check your apache error log regularly to make sure MaxClients is not reached. One easy way to keep track of this is to use a log monitoring tool, such as logcheck, to check for the appearance of the log entry. It can send you an e-mail notifying you of the situation, allowing you to increase MaxClients as needed.