CPU use in Step2

[fmurgia]

I'm running Step2 using this code:

step2_SPAtests.R \ --bgenFile=chr6_v3.bgen \ --bgenFileIndex=chr6_v3.bge.bgi \ --minInfo=0.3 \ --minMAC=5 \ --sampleFile=chr6.sample \ --GMMATmodelFile=step1_Pheno.rda \ --varianceRatioFile=step1_Pheno.varianceRatio.txt \ --SAIGEOutputFile=Pheno.SAIGE.chr6.txt \ --numLinesOutput=20000 \ --IsOutputAFinCaseCtrl=TRUE \ --LOCO=FALSE \ --IsDropMissingDosages=TRUE

It seems that this step uses all cpu available in my server (64). Is there a way to select the number of cpu to use in step2? I'm using Saige v 0.43.3 installed using conda.

[humanpaingeneticslab]

Also, notice all the time spent as "kernel" (red bars) ! Why the need for all these CPU cycles in the OS?

[weizhouUMICH]

Hi @fmurgia,

I think the multithreading has something to do with the R installation in your cluster. Then R is installed with openblas https://cran.r-project.org/doc/manuals/r-devel/R-admin.html#OpenBLAS, it will automatically use multiple threads even for simple linear algebra operation. I've found this post. https://stackoverflow.com/questions/45794290/in-r-how-to-control-multi-threading-in-blas-parallel-matrix-product You may try to follow to control the number of threads in R.

Thanks!, Wei

[fmurgia]

Hi Wei, thanks for your help. It makes sense.

Best,

Federico

[weizhouUMICH]

Hi Federico,

Thanks for the question! Here is a command that can be used to force the step 2 job to only use one CPU export MKL_NUM_THREADS=1; export MKL_DYNAMIC=false; export OMP_NUM_THREADS=1; export OMP_DYNAMIC=false;

https://github.com/FINNGEN/saige-pipelines/blob/198179aa214ffc46b6bd6a679d684ef228b1d7ac/wdl/gwas/saige_sub.wdl#L22

Actually, using multiple threads slows down the Step 2. Adding this command solves the problem

Best, Wei