PixInsight Benchmark

[fulatoro]

Late to this discussion. One way to do this is to use a mechanism that has been present in the kernel for a while called cgroups.   These are used heavily in HPC applications to solve this exact problem of NUMA caused memory bottlenecks (QPI links, cache coherence protocol issues, etc...). They essentially allow you to create a control group for CPU and Memory resources where you specify the list of CPUs and nodes associated with the group. You can then assign PIDs to the group.  I realize there was some concern about the new PID being created by PIxInsight, but my guess is that all these PIDs are children of top level thread created by PixInsight.

One way to do this though, would be to simply create the cgroup, start a shell and assign it to that cgroup and launch PixInsight from the shell.

More details on cgroups here:

https://www.kernel.org/doc/Documentation/cgroups/cpusets.txt

[vicent_peris]

Hi,

Thank you very much for your reply! Unfortunately I'm leaving tomorrow for 10 days. I'll try it in my workstation when I get back home. Meanwhile, I'll try this (if it makes sense) in my laptop.

Best regards,
Vicent.

[fulatoro]

You can experiment with the mechanisms of creating CPU/memory resources and verifying that you see the expected behavior. But obviously your laptop not being a numa system you will not see get to fully test it. The mechanisms will be the same.

Moussa

[fulatoro]

I have noticed that BPP seems to load files single threaded.  Is that correct? If so, any plans to increase read throughput by launching multiple threads to really take advantage of fast io devices? Writing is less of an issue as it is managed by the OS via the page cache (unless you start doing direct io).

Anyhow seeing that my pcie ssd is bored...