Deep performance analysis

In Part III you learned to split load across four resources and read vmstat, iostat, free. Now you how it works: not "the disk is busy" but "which system call, in which line of code, is holding the disk". For that you need tools that see inside the kernel.

The order matters more than the set of commands. First the method: it keeps you from missing a resource and from guessing. Then the tools, each for its own layer.

The USE method walks a list: for each resource you check three things - utilization (how busy), saturation (is there a queue beyond capacity), errors. It is a frame that concrete commands drop into:

CPU      util: vmstat us+sy    sat: vmstat r > cores    err: dmesg, MCE
memory   util: free available  sat: vm