Talks

Nicholas Thorne, TACC

Teach HPC software stack.
Small groups or individuals
Hardware – hard to find such a resource for teaching system admins.
Don’t change tool stacks too often
Planning for scale
Parallel file systems important, but hard to get training resources.
Training is viewed as good

Bryan Johnston, South Africa

Use older HPC systems. E.g. Ranger, Stampede from TACC, Cambridge, etc.
Puppet, Cobbler, etc. but now using in-house
Now OpenHPC [This might be a good option instead of TrinityX]

? Pawsey, Australia

Difficult to hire people, not enough timer to mentor.
Have to recruit internationally
‘HPC Sys Admin Factory’. Flexible learning paths, internal courses, exporting system admins.

Gin Tan, Monash

Asked to become a system admin of HPC from a background of non-HPC
Team has groewn

Anton Limbo, Nambia

Was asked to do HPC because new Linux in 2016.
1 week training, plus another week.
“HPC for Dummies”
Help from group in South Africa above.
Improve community support seen as important.

NSTDA, Thailand

0.2 PFlops
Worked with partners, e.g. Luxembourg, Taiwan, and learn best practices
Internships (3 months)
Need to build a career path to attract people.

Types of Admin

HPC Storage Engineer – new path?
- Maintaining, capacity planning, infrastructure, performance tuning.
- More than just system admin.
Cluster engineer
- scheduling.
Network engineer also.

Craig Morris, Edinburgh

Unique skills
Windows also possible?
FPGA, GPU, and other technologies
Build versus buy
In-house/outsourced
Technical vs Management (IT management like PRINCE, ITIL)

Perdue and SIGHPC

trying to determine scope required for training (cf. HPC-SIG’s concerns), subject matter, resources, gaps.
Longer term things like content review (again cf. HPC-SIG’s concerns)

Discussion

How to disseminate things?
Christine Kitchen’s comments. Tier 2 funding: not successful but there is an interest in funding. Need to be technology neutral but need vendor involvement to obtain sufficient resources. Issues with managerial versus technical.
Bringing industry into the mix seen as mission critical.
Student cluster challenge and links to skills?
Checking HPC system meets specification – is that a system administration job – is the engineer not the administrator. [cf. approach at Loughborough – two part verification of specification – vendor and rerun by system admins]
How do national labs do this? Yes, worth looking at this.
Bruno Silva: not making it clear what a great job it is. Incredibly hard to recruit people. Need to show the world how interesting and valuable it is to create more people interested in.
- Noted: system admins get all the complaints, none of the glory.
We know what the skills are, but need to advertise the job better
Concept of HPC not really well known?
Salary an issue – higher in industry.
Mentoring? This needs time which is often not given.
Making it easier to do the job? Share information – how easy? Avoid dealing with the same issues. Not wanting to share. [My comment would be that sharing takes time that may not be available – e.g. sharing an automation framework].
Where does HPC training fit in a curriculum?
Many online courses [I will add some links here]
Training in cloud – doesn’t sufficiently translate to the real world?
System admins don’t like to admit making mistakes. Like to solve own problems. Need to communicate problems and solutions more. Make links with other centres, bounce ideas of each other.
Teaching knowledge versus teaching how to troubleshoot.
Straw poll: 50% in audience from Comp Sci background
Breaking things important to finding out how to fix one.
Sharing [e.g. procurement docs, acceptance testing frameworks, automations – I need to do more on this in the next couple of months]
Make opportunities for HPC development clear at an undergraduate level.
- Student clusters?
Show the HPC in the science.
Look at other organisations (will add links to PERC and CARC).
There are many support tools out there: use them.

Slack: https://bit.ly/37qkoCa