Flowers are sorted by decreasing size when doing the parallel loops in BAR and Reference phases so longer jobs get queued first. But the size is measured by flower_getCapNumber() and on a small test of chrI of the yeast pangenome, the 3 jobs that stand out as longest running don't have the most caps. But they do have the longest flower_getTotalBaseLength().
Granted, this isn't a ton to go on, but this PR tries switching up the sort to use base length instead of caps. And because base length is more involved to compute, it's done in parallel.
The idea being to use this branch to run a couple big tests to see if it helps with wall time for some big, parallel, jobs. It's kind of a long shot, but runtime does seem to be bounded by a single long job for many cases, so anything to make sure it gets queued right away should lead to a measurable gain...
Flowers are sorted by decreasing size when doing the parallel loops in BAR and Reference phases so longer jobs get queued first. But the size is measured by
flower_getCapNumber()and on a small test of chrI of the yeast pangenome, the 3 jobs that stand out as longest running don't have the most caps. But they do have the longestflower_getTotalBaseLength().Granted, this isn't a ton to go on, but this PR tries switching up the sort to use base length instead of caps. And because base length is more involved to compute, it's done in parallel.
The idea being to use this branch to run a couple big tests to see if it helps with wall time for some big, parallel, jobs. It's kind of a long shot, but runtime does seem to be bounded by a single long job for many cases, so anything to make sure it gets queued right away should lead to a measurable gain...