load average

On UNIX-like systems, a positive floating-point number that expresses how busy a the CPU is. Theoretically, a value of 0 is completely idle, a value of 1 is the maximum work that a single-processor computer should do, and anything higher than that is the product of some sort of incantation calculating how abusive you are to your server. Load averages are most commonly seen in the output of the uptime command.

Example:

$ uptime
 12:34am  up 51 days,  6:12,  1 user,  load average: 7.62, 7.64, 6.68

(Yes, I was mean to that server... Running 3 mp3 encoders and a seti@home is not good for a pentium 75. At least I have it on a UPS, so it's uptime is way better than my workstations...)

The load average given by Unixen expresses, basically, a percentage of how much CPU time is being utilized by running processes. There are many processes (such as init) which rarely ever need CPU time. When all processes are in this wait state it is quite possible to have a load average of 0.00.

The ideal situation is 1.00, though. This means that every CPU cycle is used, but no processes are queued, waiting for time when they would normally be receiving it. You might get this when running SETI@Home on an otherwise-idle box, because SETI@Home will use additional CPU cycles only until the load is 1.00.

When the load exceeds 1.00, you have all CPU cycles in use, and processes queued waiting for time, rather than getting it right when they need it. This is when CPU lag occurs. A load of 2.00 means there are just as many processes are waiting for CPU time as there are processes receiving the time they need. 3.00 means there are twice as many waiting as there are running. 4.00 means three times as many, and so on.

While pretty close, most descriptions of load average misreport one essential piece of information. Load average is the average number of processes either in the running state (often marked with an R in programs like top) or in the waiting for disk state (often marked with D). That means that even if your system is hardly using any CPU but several processes are reading from disk your load can still be quite high. This makes sense, as it's an indication of system load, not just CPU load.
Note that most unixes treat NFS and other network filesystems as disk, so when you unplug your laptop while reading files over NFS, your load will spike because every process trying to access the now unreachable files will add one to the load average.

Load average is usually displayed as three numbers, the first being averaged over the last minute, the second over the last 5 minutes, and the third over the last 15 minutes.