Is there any speed benefit to redirecting Djava.io.tmpdir to local scratch?
I'm currently putting together a workflow for WGS (Illumina 30x) variant calling (HaplotypeCaller) using GATK 3.2 on our local SGE grid. I was wondering whether there is any speedup with redirecting -Djava.io.tmpdir to scratch space on the local node? Which steps are I/O intensive and could potentially benefit from this? Currently the bottleneck seems to be IndelRealigner which takes ~14hrs.