3.3 KiB
SIG/HPC meeting 2023-04-20
Attendees:
* Alan Marshall
* Nje
* Neil Hanlon
* Matt Bidwell
* David (NezSez)
* Jonathan Andreson
* Stack
* Balaji
* Sherif
* Gregorgy Kurzer
* David DeBonis
Quick round of introduction
Everyone introduced themselves
Definition of stakeholders
"still needs lots to clarification and classification since those are very wide terms"
* HPC End-user ?maybe?
* HPC Systems admins and engineers, to provide them with tools and know how to build HPC clusters using Rocky linux
* HPC Vendors, however the SIG has to be vendor neutral and agnostic
Discussions:
Stack: we need to make sure that we are not redoing efforts that already done with other groups Greg engaged with Open HPC community and providing some core packages such as apptainer, mpi, openHPC
Sherif: we need to have one hat to fit most of all but we can't have one hat that fit all Stack: Feedback regarding Sherif's idea that generic idea's are not great idea and there is a bad performance Greg: we need to put building blocks in the this repo and will make life easiest and lower the barriers like Spack, slurm and easybuild
Devid (NezSez): Some end users won't understand / know anything about HPC and just needs to use the HPC, such as Maya or dynamic fluids
Neil: some tools can be very easily an entry point for organization and teams to use HPC like jupiter playbook
Stack: HPC is usually tuned to different needs, we can reach to other HPC that are running Rocky to ask them to promate rocky and establish a dialog to get an idea of what things that they are running into rocky
Matt: HPC out of the box there are few projects that doing that and we don't need to run in circles of what we are going to
Balaji: SIG for scientific application that focus on support the application and optimization, and HPC suggest the architecture to reach max capabilities
Greg: Agreeing with stack we don't want to provide application that there are tools that do that
Gregory Kurtzer (Chat): A simple strategy might be just to start assembling a list of packages we want to include as part of SIG/HPC, and be open minded as this list expands.
Neil Hanlon(Chat): actually have to leave now, but, if we make some sort of toolkit, it has to be quite unopinionated... OpenStack-Ansible is a good example of being unopinionated about how you run your openstack cluster(s), but give you all the tools to customize and tune to your unique situation, too
Remarks:
* A point raised, should be rebuild some packages that area already in Epel or not and if we shall have a higher priority on our repo or not
* We need to think more about conflicts with other SIGs like lustre and sig storage
Action items:
* List of applications “Thread on MM to post pkgs”
* Building blocks which are each pkg as a building block such as lustre, openHPC, slurm, etc…
* Reach out to other communities “Greg”
* Reaching out for different sites that uses Rocky for HPC “Stack will ping few of them and others as well -Group effort-”
* Reaching out to hardware vendors
* Statistic / public registry for sites / HPC to add themselves if they want
* Meeting will be bi-weekly “Tantive Thursday 9:00PM UTC”
* Documentations