Recognizing the Efficiency Benefits of CPU Threading
I've never met anyone who actually believes these measures offer an efficiency benefit; that is, until they test it themselves. (Warning: It takes some proficiency with AIX numbers to be able to recognize the efficiency benefit.)
Anyway, here's what you need to learn, know, ask yourself and ultimately do to accomplish this in your environment:
1a) Understand the runqueue thread count relative to the total count of logical CPUs.
1b) Monitor the value of AIX:vmstat -IWw 1:kthr:r until familiar.
2a) Understand how CPU idle (AIX:vmstat:cpu:id) and CPU wait (AIX:vmstat:cpu:wa) are calculated.
2b) Monitor the values of AIX:vmstat -IWw 1:cpu:id and cpu:wa until familiar.
3a) Understand the meaning of the AIX:vmstat -IWw 1:cpu:pc and :cpu:ec values.
3b) Monitor the values of AIX:vmstat -IWw 1:cpu:pc and :cpu:ec until familiar.
4a) Are there mostly vCPUs exhibiting ST/SMT-1 threading? I expect yes.
4b) Monitor AIX:mpstat –w 2 for general SMT-1, SMT-2 and SMT-4 threading patterns.
4c) Note the pattern of vCPUs in SMT-1, SMT-2 and SMT-4 threading mode.
5a) Remove a vCPU and monitor the change in AIX:mpstat -w 2 threading patterns.
5b) Remove vCPUs until vCPUs are generally in SMT-2/SMT-4 threading mode.
6a) Note the values of AIX:vmstat -IWw 1:cpu:idle and cpu:wa to realize the efficiency benefit.
6b) Note the values of AIX:vmstat -IWw 1:cpu:pc and :cpu:ec to realize the efficiency benefit.
These instructions are an example of tactical monitoring by the numbers. It is the basis of POWER/AIX tuning, which is a skill that takes practice. Obviously you're monitoring your LPARs already, but the intent with this series of articles is to help you conduct more meaningful monitoring. Starting with part three, I'll detail and illustrate the meaning and use of these steps. I hope you'll stay tuned.
Like what you just read? To receive technical tips and articles directly in your inbox twice per month, sign up for the EXTRA e-newsletter here.