Scaling studies

[GayatriManda]

Review Comments on Scaling Study Episode

1. In the questions section, it may be useful to add another motivating question such as:

   • “Will the application benefit from requesting more computational resources?”

2. In the What is Scaling section:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L40

   the text mainly discusses execution time scaling with increasing processors, but it does not explicitly discuss the role of problem size. As currently written, it indirectly describes strong scaling without clearly distinguishing it from weak scaling. Since Amdahl’s law and Gustafson’s law are introduced later, this may confuse readers. The section could benefit from clearer wording.

3. In this line:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L110

   grep is mentioned, but the actual command is not shown there. The command only appears later in the episode:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L165

   It may improve readability to either introduce the command earlier.

4. In the callout section on “Overhead and reliable measurements”:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L194

   it may also be useful to briefly discuss communication overhead, especially since it is mentioned later here:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L220

5. Regarding these lines:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L273-L275

   apart from the showing the plot, it may be interesting to discuss possible reasons why the speedup drops from 0.91 → 0.78 → 0.55. This could naturally connect to a short discussion on identifying bottlenecks in later episodes.

6. In this line:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L340

   I believe the formula may contain a typo and should probably be:
   s + N(1 - s)

7. In this section:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L373

   it may be helpful to emphasize that identifying which application parameters affect runtime is important before designing scaling studies. This could then naturally lead into examples such as -spp in the discussed scenario.

8. In this paragraph:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L396-L400

   the figures are referred to as “first” and “second” image. It may be clearer to use more descriptive names or captions. Also, the statement suggesting that increasing the number of pixels is useful may be somewhat debatable, since usefulness depends on the scientific goal (e.g., higher resolution vs reduced image noise).

9. In this line:
   https://github.com/carpentries-incubator/hpc-job-efficiency/blob/main/episodes/04_scalingstudy.md?plain=1#L406

   it is stated that the application is “weakly scaling,” but the intended meaning was not entirely clear to me.

10. In the final weak scaling challenge, the “speedup” and “efficiency” columns appear to be empty in both the exercise and the solution sections.