mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   Linux (https://www.mersenneforum.org/forumdisplay.php?f=39)
-   -   Running mprime in a Kubernetes cluster (https://www.mersenneforum.org/showthread.php?t=25407)

brannondorsey 2020-03-29 19:07

Running mprime in a Kubernetes cluster
 
Hi folks,


Just wanted to share a little project I've been working on recently. I've packaged mprime as a Helm chart, so that it can be easily deployed in a Kubernetes cluster and run as a low priority workload to donate idle compute.



You can find the package on the new-ish Helm Hub distributed package search: [url]https://hub.helm.sh/charts/brannon/mprime[/url]


And the source code is available here: [url]https://github.com/brannondorsey/helm-charts/[/url]

I've been running it on a hobby cluster for a few weeks and had great results. Its great for shorter tasks, like trial factoring, as mprime containers can be configured to be preempted when higher priority workloads are required in the cluster.


I'd be curious if anyone else is running K8s clusters or uses Helm.



P.S. First post here, so please go easy on me . Not sure if this is the right subforum for this post. Happy to move it or continue the conversation elsewhere if that's preferred.

Uncwilly 2020-03-30 00:17

Not sure how you have it all configure, but there is at least 1 important factor to consider with clusters.
For the best total throughput, don't run a single test on more than a sing;e chip. Once you start using more than one there is a great drop off. If you are only trying to test one number as fast as possible, more than one chip can be used. But if you want to do as many tests as possible in a given period, keep each test on a single chip.

kriesel 2020-04-22 21:10

[QUOTE=brannondorsey;541256]I've been running it on a hobby cluster for a few weeks and had great results. Its great for shorter tasks, like trial factoring, as mprime containers can be configured to be preempted when higher priority workloads are required in the cluster.
[/QUOTE]Just a sort of FYI, mprime implies cpu use, not gpu use, and TF is better done on gpus, with P-1 and primality testing on cpus. P-1 is quicker than primality testing, but takes more memory.


All times are UTC. The time now is 10:01.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.