"Results of the survey for HPC practitioners on the HPC Training Community"

So far we received 55 submissions for our survey. We conducted an intermediate analysis of the submissions. We thank our contributors from HPC CF and the NHR centers.

The questions covered three sections:

  • Satisfaction
  • Training experience
  • Personal information

The personal information was only collected for statistical purpose.

Results

Satisfaction

Most participants indicated that their systems are rather easy to use (85% at least average to high satisfaction). 7 Attendees were not satisfied. Also, most attendees were pleased when preparing and executing apps on their HPC system. Individual comments for improvement contain two major themes:

  • Difficulty to set up and retain a working environment including dependency handling and packaging, missing standard system software, behavior after system tools - it is easier to use a laptop

  • Providing more interactive tools such as Jupyter Notebooks   Documentation was similarly assessed, 5 participants marked less than average. For training courses provided by the HPC community, 6 participants were less satisfied but 32% were just average. Individual comments for satisfaction with training include various positive feedback. Room for improvement cover:

  • better preparation of trainers
  • some detail or important topics are lacking
  • the goals and coverage of training is not well defined

In the opinion of participants, a data center could support them better by:

  • more helpdesk interactions and direct contact as a ticket system isn’t optimal due to delayed response time
  • providing more training, potentially mini-courses (e.g., write in bash) and more experienced trainers
  • providing best practices and rules of thumb for running good and efficient jobs

Training Experience

There had been various categories that had been investigated.

  • Most attendees found they had attended sufficiently many pieces of training.
  • At the same time, 51% indicate they would need more training to do their job productively.
  • About 20% miss information on what is going on under the hood of the system.
  • A small fraction has no time for proper training and misses important topics.
  • Finding the schedules of training is easy and there is no trouble registering.
  • Some struggle with costs for training though.
  • About 30% feels training is not recognized among professionals and difficult to identify which trainings are important to them.
  • 60% would appreciate a certificate for training.
  • About 50% prefer virtual workshops over in-person events.

Among the most pressing issues related to training were:

  • Industry recognition of training
  • Lack of hands-on-training
  • Intermediate training is missing
  • Interacting effectively with an HPC system from a user-friendly environment
  • Clear indication of goals and takeaways of students

Personal information

About 45% of attendees attended less than 2 training events over the last 5 years while 40% attended between 2 and 5 events and 10% less than 10 events.

The topics the attendees are most interested in were Performance Engineering (60%) and Big Data Analytics (51%) followed by using the HPC environment and HPC fundamentals (49%).

The survey is not fully representative, in the cohort, 74% of the attendees were from Germany and 48% PhD students followed by 26% of researchers. Most have been using the system for between 1-5 years while 34% are using HPC for less than a year.

Conclusions

We believe that the HPC-CF efforts will address some of the issues mentioned by the attendees in the free text forms: by labeling a training with the skills - it will be clear what the content will be and the recognition of experts and industry are addressed by the certificating itself.

Data

As promised and indicated in the survey, we release the raw data of the submitters. The non-textual answers are published in 2021-training-community-responses.csv.