Part 1 of 7 · 2 min read
Why this page comes fourth
In this module you will
- State the USE method: for each resource, check utilisation, saturation and errors.
- Read
vmstatcolumns (r,b,si,so,bi,bo,us,sy,id,wa) and say what each tells. - Tell a processor-bound workload from a disk-bound one by running both and comparing the numbers.
- Read the load average against the number of cores, and say why it lags.
- Follow a decision table from a symptom to the resource that is the bottleneck.
Page 1 gave the tools, page 2 the files behind them and page 3 the view of system calls. This page puts numbers to work: a method that says which numbers to read and in which order. It uses the scheduler of Level 3, the memory counters of Levels 6 and 7 and the disk model of Level 8, including the note that a process in uninterruptible sleep counts in the load. Page 5 then covers what happens when a program fails and not only when it is slow.
About this example About this example
Real: Both C programs were compiled with gcc -Wall -Wextra and run on a Linux test machine (kernel 6.18, 2 processors, 8 GiB, no swap). All vmstat, ps and load figures are real output taken while the programs ran.
Scripted: The programs burn (an endless loop) and syncer (a write with fsync in a loop) are made up to give a clean processor-bound and a clean disk-bound workload. The 4-hog test uses twice as many hogs as cores on purpose.
Not real: The test machine's disk is a fast virtual one, so the disk-bound case shows a moderate wa and not the long waits of a spinning disk (Level 8). The memory case is described from the counters and not run, since the machine has no swap.