Produce and deplay a questionnaire of community training needs - Task 013

Work Package: wp2.3
Open

Fit to programme

This task has been identified by the working groups as part of the agenda behind WP 2.3 and WP 2.4.

The task number is 013.

Description

A plethora of training already exists relating to accelerated compute (the collation of which is the task of WP2.1). For example, writing vector addition code for GPUs has dozens of different tutorials available. There are however many areas where it is less clear how someone currently not skilled in the art may become a competent practitioner.

It is not feasible for us to identify every possible gap; therefore, we need to survey the community to build a picture of where there are opportunities for improvement, and what the relative demand is in each case. There are at least three communities that the survey must be inclusive of:

  1. Research Software Engineers: “If I wanted to write an Ising model implementation on a Cerebras Wafer Scale Engine or Graphcore IPU, how would I do that?”
  2. Research Infrastructure Engineer: If I want to set up a GPU cluster software stack from scratch such that Slurm will give best performance, how do I learn what to do?
  3. Researcher software/infrastructure user: How should I pin processes to get best performance out of my software on each cluster I use?

In each case, the survey and advertisement of it must encourage participation not just from those with the specified job titles, but also researchers and academics who have such responsibilities as part of their work who would not identify with the labels. For example, a postdoc who writes software, or a junior academic who has bought a small cluster with grant funding.

The survey must be advertised widely to maximise participation. It may be, for example, advertised at CIUK, potentially with preliminary/early findings, to encourage more participation

Outcomes

  • A survey design of a survey on training gaps in accelerated compute
  • The dataset resulting from running the survey, with sufficient results from each of the communities discussed above
  • The workflow used to analyse the dataset
  • A report on the results of the analysis, including preliminary recommendations of training topics to develop further
Apply now