Not really related to your problem exactly, but we'e needed to remoe the optical dup counting method as well in some cases.
In mouse (and other species) it sometimes happened that we had a few. 1-2, hotspots where there were A LOT of dups (>800,000). The double loop in findOptical dup would run for a long, long time. Remoing this loop )and losing optical duplicates) would make Mark dup run in minutes, as you mentioned. Louis On 14-08-21 08:02 PM, Thomas Sibley wrote: > On Aug 21, 2014, at 15:29 , Thomas Sibley <[email protected]> wrote: >> On Aug 21, 2014, at 15:10 , Thomas Sibley <[email protected]> wrote: >>> The runtime is still dramatically different from my other input, and I >>> wonder if you have any ideas why? Do you expect it's some property of the >>> data? Or some pathological edge case? >>> >>> I can't help but think the crash and runtime leap are related... >> >> Bingo. After putting together my jstack/strace observations with a reading >> of the MarkDuplicates source, I wrote the one-line patch attached, >> recompiled MarkDuplicates.jar, and the runtime dropped to 2-3 minutes. I >> started a tight loop running and re-running MarkDuplicates on the same >> problematic input and will see if I can provoke a JVM crash still. > > 100 runs later, doing 4 at a time, and I haven't provoked a JVM crash with my > patched MarkDuplicates. I also tried with the IntelDeflater re-enabled and > did not trigger a crash. Average run times of 1.38 minutes. > > _Something_ is bogging down in findOpticalDuplicates and going off the deep > end. I'd guess the Collection.sort call since I saw mention of the sort > implementation guts in the crash logs, but again, I'm no expert on the JVM. > > Thomas > ------------------------------------------------------------------------------ > Slashdot TV. > Video for Nerds. Stuff that matters. > http://tv.slashdot.org/ > _______________________________________________ > Samtools-help mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/samtools-help > ------------------------------------------------------------------------------ Slashdot TV. Video for Nerds. Stuff that matters. http://tv.slashdot.org/ _______________________________________________ Samtools-help mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/samtools-help
