From bdc229efc6ed8210d1b8af1b2bb128e2a08a9c32 Mon Sep 17 00:00:00 2001 From: attesillanpaa Date: Wed, 12 Feb 2025 17:02:40 +0200 Subject: [PATCH] extend scaling test exercise added columns for slurm jobids and a note that encourages preparing a report with all requested details to qualify for (asking for) study credits. --- part-2/workflows/tune-performance.md | 33 +++++++++++++++++----------- 1 file changed, 20 insertions(+), 13 deletions(-) diff --git a/part-2/workflows/tune-performance.md b/part-2/workflows/tune-performance.md index 5d425d4d..3f695185 100644 --- a/part-2/workflows/tune-performance.md +++ b/part-2/workflows/tune-performance.md @@ -82,12 +82,12 @@ example simulation. Calculate the speedup by dividing the previous elapsed time with the elapsed time obtained using twice as many nodes: - | Number of nodes | Elapsed time (s) | Speedup | - |:---------------:|:----------------:|:-------------------------------:| - |1 | | - | - |2 | | *t*1/*t*2 | - |4 | | *t*2/*t*4 | - |8 | | *t*4/*t*8 | + | Number of nodes | Elapsed time (s) | Speedup | SLURM JobID | + |:---------------:|:----------------:|:-------------------------------:|:---------------:| + |1 | | - | | + |2 | | *t*1/*t*2 | | + |4 | | *t*2/*t*4 | | + |8 | | *t*4/*t*8 | | ☝🏻 Remember that the speedup should be *at least* 1.5x when you double the number of cores! This is important to ensure that the CPU resources are used @@ -136,13 +136,13 @@ environment variable to be equal to the number of threads per task. 3. Complete the table below: - | MPI tasks per node | OpenMP threads per task | Elapsed time (s) | Memory utilized (GB) | - |:-------------------:|:-----------------------:|:----------------:|:--------------------:| - |128 | | | | - |64 | | | | - |32 | | | | - |16 | | | | - |8 | | | | + | MPI tasks per node | OpenMP threads per task | Elapsed time (s) | Memory utilized (GB) | SLURM JobID | + |:-------------------:|:-----------------------:|:----------------:|:--------------------:|:-----------:| + |128 | | | | | + |64 | | | | | + |32 | | | | | + |16 | | | | | + |8 | | | | | 💭 Were you able to run the calculation faster by launching multiple OpenMP threads per MPI task? What is the optimum ratio? @@ -151,6 +151,13 @@ threads per MPI task? What is the optimum ratio? task? Use the `seff` command to check. Can you explain the reason for your observation? +Note: if you plan to apply for study credits for this course, prepare a report +including the tables above and discussion on all questions with the 💭 symbol +and (upload the report and) present it with the course certificate to the local +authority granting credits. CSC cannot grant credits but for carefully and +correct reports we recommend granting them. + + ## More information - [Docs CSC: Performance checklist](https://docs.csc.fi/computing/running/performance-checklist/)