CPU Usage Stats – Monitor Core Usage with Top

Two dual-core processors contain 4 CPUs. Multi-threaded processors also exist. Each thread appears as a CPU to the OS, although they are not true CPUs as are cores. Most usually multi-threaded CPUs have two threads per core, so you can have a single, dual-threaded, dual-core processor that appears to your OS at 4 CPUs.

It should be noted that Sun Microsystems had a T2 chip which had 8 cores and 8 threads per core for 64 virtual CPUs on a single processor.

Percentage of cpu is reported differently, by different tools, by different systems. A better way to consider cpu load is with load. Consider the below overloaded worker machine:

# w 
 02:22:31 up 221 days, 11:06,  1 user,  load average: 9.87, 9.50, 7.25
USER     TTY      FROM              LOGIN@   IDLE   JCPU   PCPU WHAT
stephan  pts/0    173.13.169.18    02:22    0.00s  0.44s  0.00s w


~$ cat /proc/cpuinfo |grep processor
processor   : 0
processor   : 1

This says I have a 1 min load of 9.87, 5 min load of 9.50, and 15 min load of 7.25. The 'load' number represents how many processors worth of work this machine has been assigned, while the cpuinfo command shows me how many actual processors I have to do the work. If I had 12 cpus, this load level would be perfectly fine.

PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND                                                                                
11579 app       20   0  263m  97m 4104 R   22  1.3   0:00.85 ruby                                                                                   
11586 app       20   0     0    0    0 Z   20  0.0   0:00.62 ruby <defunct>                                                                         
11589 app       20   0  262m  96m 3884 S   18  1.3   0:00.53 ruby                                                                                   
11592 app       20   0  260m  95m 3000 R   17  1.3   0:00.50 ruby                                                                                   
11600 app       20   0  260m  95m 2744 R   15  1.3   0:00.45 ruby                                                                                   
11595 app       20   0  260m  95m 2744 R   13  1.3   0:00.39 ruby                                                                                   
11598 app       20   0  262m  95m 3096 R   12  1.3   0:00.35 ruby                                                                                   
11604 app       20   0  258m  93m 2744 R   10  1.3   0:00.30 ruby                                                                                   
11607 app       20   0  257m  92m 2496 R    8  1.2   0:00.25 ruby                                                                                   
11610 app       20   0  256m  91m 2560 S    4  1.2   0:00.11 ruby

So you can see cpus are being divvied out between processes, but what I care is that there's more work than the cpus can feasibly keep up with. That's causing queued jobs to have to wait for cpu to be free to use them.