Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Understanding CPU on AIX Power SMT Systems (ardentperf.com)
43 points by luu on Feb 17, 2022 | hide | past | favorite | 6 comments


A few changes to note since the P7/P8 days. As of P9, the default is now SMT8 instead of SMT4.

Changing the profile from say P8 to P9_base to P9 changes things under the hood in terms of processor capabilities. This roughly translates into VM hardware versions in VMware-speak.

There are cores and then there are virtual processors as well as the notion of processor folding. Here's a starting point: https://www.ibm.com/support/pages/aix-virtual-processor-fold...

Some P9 specific changes re folding: https://www.ibm.com/downloads/cas/39XWR7YM

Source: I have to deal with POWER systems day to day.


> There are cores and then there are virtual processors as well as the notion of processor folding.

I'm really not sure if I understood that "processor folding" correctly -> would that translate in the case of Intel/AMD/Linux to something like a "dynamic hyperthreading" done under-the-hood by the OS?


I would say that folding is not a consequence of SMT, but of the assignment between the VP (virtual processors) assigned to the LPAR (VM) and the processing units assigned. So more a effect resulting from the virtualization with consequences in conjunction with SMT.


Furthermore, the folding behavior can be adjusted. From the second link that I posted:

In addition to the reduction of VPs, you can consider setting the schedo tunable “vpm_throughput_mode” to a value of 2. By default, running in raw throughput mode (i.e. vpm_throughput_mode=0) AIX spreads all available work across as many VPs as available, dispatching work first to the primary SMT thread of each VP, then the secondary SMT threads, and so on, thus providing the best performance in most workloads. This configuration, however, can lead to a higher PC (processor capacity) utilized by the LPAR. Depending the workload, you can reduce the PC by setting the vpm_throughput_mode tunable to 2, thus having AIX to schedule work to the primary and secondary threads equally.


Seems as good a time as any to ask about the current purpose and future of AIX, HP-UX, and similar. They seem similar to mainframes now. I feel like they only exist to support legacy systems -- even large enterprises today really seem to want to horizontally scale on Linux.


AIX is usually deployed because there is a requirement for DB2, from my experience. You can use it as a clustered hypervisor too, the tooling is fairly decent. It is a LOT better than HPUX (which I think has always been absolutely terrible) but not as good as Solaris.

It was already “dead” when I first worked on it more than a decade ago. I think it’s a lot easier to replace with Linux than mainframes too. Doesn’t have a really good moat.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: