search

cancerit / PCAP-core

NGS reference implementations and helper code for mapping (originally part of ICGC-TCGA-PanCancer)
GNU General Public License v2.0
9 stars 10 forks source link

map_threads should use all cores if possible #41

Closed jsmedmar closed 5 years ago

jsmedmar commented 5 years ago

If you use -t 32 and you have only one input (e.g. aligning from one bam) - bwa_mem.pl will just use 6 cores. Instead it should be using 32.

keiranmraine commented 5 years ago

Even with a single BAM as input the -fragment option is in use, you could get many 6 cpu instances running in parallel on the same BAM input.

If you have set the fragment value high to prevent splitting you can specify -map_threads to be equal. You will not get linear performance improvement and for values >12 you should also investigate this option:

    -bwa_pl      -l   BWA runs ~8% quicker when using the tcmalloc library from
                      https://github.com/gperftools/ (assuming number of cores not exceeded)
                      If available specify the path to 'gperftools/lib/libtcmalloc_minimal.so'.