jzluo
New Member
Posts: 1
|
Post by jzluo on Jun 14, 2017 20:19:04 GMT
Hello, My MLMA is only running on one thread for some reason. I think everything up to the association analyses (reading in the files, the GRM, etc) is running on multiple threads, but only one thread is used when it gets to the association analyses. Is this normal? example: "Options: --mlma --grm allgrm05 --bfile test --mlma-subtract-grm fz60k_chr1 --chr 1 --pheno test.phen --covar covar05.covar --qcovar qcovar05.qcovar --out test_loco_chr1 --thread-num 30
Note: the program will be running on 30 threads."
Thank you!
|
|
|
Post by zhilizheng on Jun 27, 2017 8:21:25 GMT
Hi,
Thank you for your question. I'm investigating on this.
I find the bottleneck of MLMA in first glance. It might be calcu_Vi function.
We have improved another function calcu_Vi_bivar (for GREML bivar) for the next planned release. We will improve this part in next few days.
We are sorry that you might need to wait until we release the next version to solve this problem. As we update the underlying packages for performance, further test are needed to make sure it works as before.
Cheers, Zhili
|
|
|
Post by zhilizheng on Jul 14, 2017 2:10:04 GMT
Hi jzluo,
I have checked the code of MLMA. There are some blocks of matrix calculation that are not paralleled. It might be due to your large data, which caused the running on these blocks for a long time. So, you observed the program only run in one thread.
I have make some revisions in the codes, paralleled these blocks. We will release the new version soon.
Hopes it works this time.
Reply this post if there are still some problems after you try the new release.
Cheers, Zhili
|
|