Comments (3)
  • Add a Comment
  • Edit
  • More Actions v
  • Quarantine this Entry

1 ManojBableshwar commented Permalink

great article.. specially the part that shows that SMT3 is not possible.. AIX will adjust it as SMT4 plus SMT1 for 5 processes :-)

2 B.Eckenfels commented Permalink

Thanks for this post. If we dont talk about processes but kernel threads (think Java) how is the situation then? Will it try to keep all busy threads of a single process onto the same CPU (and locality if we think NUMA). So does it happen that a process with multiple threads get scheduled on a SMT=4 CPU before any of the other CPUs goes to SMT=2? <div>&nbsp;</div> Can you give some hints how to diag these things? Especially from within a LPAR to see how many of the logical CPUs are scheduled on which Physical core? <div>&nbsp;</div> If you set up LPAR you specify the number of CPU cores not the number of logical (host) CPUs? And the SMT setting is a property of the guest OS image? Is this somewhat depending on entitlement as well? So for ecxample if a core is shared by 2 LPAR can it be enforced that none of those LPARs use SMT&gt;1?

3 B.Eckenfels commented Permalink

Forget my last question, I should have read the whole series first :)