Monitoring and bottlenecks

When a system gets slow, there is one question: where is it stuck. In the CPU, in memory, in the disk, or in the network. Staring at top does not always show it, and that is where guessing starts. This chapter gives you the tools and an order to check resources without guessing.

Start with the number everyone sees and almost everyone reads wrong.

Load average is not CPU usage in percent. It is the average number of processes that are right now either running on a CPU, or waiting in line for one, or stuck in uninterruptible disk wait (state D). The three numbers are averages over 1, 5, and 15 minutes.

$ uptime
 14:22:01 up 9 days,  3:14,  2 users,  load average: 4.10, 2.30, 1.05

For the number to mean anything, you divide it by the nu