At NeSI we want to help you run your work as efficiently as possible on our platforms, so that you obtain your results in a timely manner and so that we can fit in as much work as possible, hopefully reducing queue times for all users.
With this aim in mind, we may ask to perform a "job efficiency audit" of your work on NeSI, particularly if you have been granted, or are requesting, a large compute allocation. See below for more details about what to expect during this process. If you are interested in us looking into your job efficiency for potential improvements, please contact us at firstname.lastname@example.org.
At the end of a job efficiency audit you could expect one of the following outcomes:
- We determine that your workflow/jobs are running efficiently on our platform
- Some areas for improvement are identified (and agreed with you)
- For "quick wins" we may be able to achieve these improvements within the scope of the job efficiency audit
- For larger pieces of work, we would assist you in applying for a NeSI Consultancy project, where we would work with you on a longer term project to implement any agreed changes
What you can expect from us
During a job efficiency audit you can expect that we will:
- Spend some time (typically up to 10-20 hours) to investigate your software and workflows that you are running on NeSI, to determine whether there is an opportunity for optimisation or efficiency improvements
- Communicate clearly and pass on any suggestions for improvements that we identify
What we expect of you
During a job efficiency audit, some input will be required from you, such as:
- Investing time to answer questions, provide code and input data as necessary and make changes to your workflow if needed (this may involve attending some Zoom meetings and/or email communication)
- Setting up some test configurations that we can use for profiling and benchmarking your jobs; these should be representative of your work but don't necessarily need to be complete calculations. For example, with a simulation code we could choose to reduce the number of time steps but keep the domain size the same.