Gentoo Forums
Gentoo Forums
Gentoo Forums
Quick Search: in
High Performance Computing (HPC) in Gentoo
View unanswered posts
View posts from last 24 hours

 
Reply to topic    Gentoo Forums Forum Index Other Things Gentoo
View previous topic :: View next topic  
Author Message
vcmota
Apprentice
Apprentice


Joined: 19 Jun 2017
Posts: 159

PostPosted: Tue May 14, 2019 3:05 pm    Post subject: High Performance Computing (HPC) in Gentoo Reply with quote

I just purchased a 32 core AMD threadripper 2990wx for using in scientific computing, and I want to configure it like a cluster. I am aware of this tutorial, but I am not sure how it is applicable to my needs. What I want is first to know if it is possible to access each of the nodes as if they where isolated nodes, in such a way I would first accomplished the following result:


  • one of the nodes of the processor would run as a master, running all the tasks of a very basic computer, in which I would make a very very basic gentoo install, the most basic and lightweight I can. It would mainly need SSH, and a way control all the other 31 nodes. I don't even plan to install X on the computer, so a graphic set up wont be needed at all. This node would not run any calculation at all, in such a way that it would not be slowed down by the calculations, and its activities such as SSH would not disrupt/interfere/slows down the calculations;
  • in the other 31 nodes the ONLY thing I want to do are my calculations, which might be of the parallel type or serial type. I don't want those nodes occupied with nothing that is not calculations;


Is this even possible? If it is, what do I need to learn/use from the gentoo tutorial?

Another possibility would be to get the whole 32 nodes as just a calculation workstation, or something like that. Than I would use another computer (that I have to spare) to access the AMD via an internal LAN. But that does not looks like the best solution, because while the calculations are running the AMD computer would be terribly slow to access, and also the results of the calculations would also be very slow to access. And the access itself would slow down the calculations. Am I right?


Thak you all.
Back to top
View user's profile Send private message
Ant P.
Watchman
Watchman


Joined: 18 Apr 2009
Posts: 6141

PostPosted: Tue May 14, 2019 11:09 pm    Post subject: Reply with quote

Yes, this is possible with some work. Look at the cgroup-v2 documentation for "cpusets" to keep groups of processes firewalled off from other CPUs, and the nohz_full kernel cmdline switch.

You'll also need to be careful with how processes on the 31 compute CPUs are allocated because TR has a NUMA layout.
Back to top
View user's profile Send private message
vcmota
Apprentice
Apprentice


Joined: 19 Jun 2017
Posts: 159

PostPosted: Wed May 15, 2019 1:59 am    Post subject: Reply with quote

Thank you Ant P. for your reply. I took a look on what you suggested and unfortunately it feels this is something way out of me league. I am afraid that I don't have the knowledge to do it by myself. I was hoping that this could be done solely via torque or some other specialized suit of codes and that there would be tutorials available.

In your opinion, if you don't mind sharing, is it a good idea in the first place? It seems so to me, because this is how many real clusters that I access are configured: one and only one of the nodes is dedicated to all the administrative tasks, to preserve all others for what really matters. So why not try something like this on a multi-core workstation?

Thank you again.
Back to top
View user's profile Send private message
bec
Apprentice
Apprentice


Joined: 30 Sep 2004
Posts: 218
Location: Cali - Colombia

PostPosted: Wed May 15, 2019 2:09 am    Post subject: Reply with quote

slurm is a job manager used in HPC, it is packaged for Gentoo:

https://packages.gentoo.org/packages/sys-cluster/slurm

I have no experience installing it, but I have used it in a Centos cluster, and the CPU usage can be configured in general and with limits per job.
_________________
abe
Back to top
View user's profile Send private message
vcmota
Apprentice
Apprentice


Joined: 19 Jun 2017
Posts: 159

PostPosted: Wed May 15, 2019 3:37 am    Post subject: Reply with quote

Thank you bec for your reply.

I was taking a look at their upstream page and it seems that slurm would help in that task indeed. I would probably need to first install the base system and after that isolate the system processes via slurm in a given CPU. The bad thing is that slurm is not in the stable tree, but I guess I would have to live with that.
Back to top
View user's profile Send private message
Display posts from previous:   
Reply to topic    Gentoo Forums Forum Index Other Things Gentoo All times are GMT
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum