LinuxQuestions.org - NUMA support in the kernel configuration in the default slackware kernel package.

Hi.

Currently, I'm a graduate student, majoring in the computer science.

My work is mainly concerned with the machine learning (especially, python and TensorFlow. Yes, the hype). I think that the de facto standard to start with this job is to install an ubuntu distribution (or debian based OSs, which provide apt-get).

(By the way, please let me make it clear that I just mention that this is the recommended way for many others, for example, all colleagues of my lab except me use ubuntu distribution for the research, but definitely I do not want any kind of aggressive debates. Please excuse me if you mind that.)

Yet, for me, being a slackware user for over twenty years since 1995, before the millennium,

(I remember that the time when I was a kid, that was, a freshman in the college, and we needed lots of floppy disks to install slackware for the 80486 computer),

so I do not want any distribution but slackware.

However, in the default configuration of the slackware kernel, what is missing is the NUMA configuration. Here's the dependency chain:

TensorFlow -> CuDNN (by NVIDIA) -> CUDA driver (by NVIDIA) -> kernel (which lacks NUMA support),

so the first thing when I set up a new machine (or after upgrading the kernel, e.g., for security update) is to compile a local version of the kernel to include CONFIG_NUMA.

I know that compiling a kernel is not a big deal, but as a matter of fact, it's time consuming and needs extra work whenever the kernel is updated, so I'll be very glad if I can use the NUMA feature from out of the box, i.e., from stock kernel of the slackware release.

I know that deciding a feature of the distribution can be very difficult, for there can be so many opinions, so after all this is just my humble opinion, but I think it will be helpful to many others for whom within the situation like me.

I'll very glad to hear about this matter.

Thank you very much.

Best regards,

Sungjin, Kim.