When loci are sufficiently dense, most of the time the genotyper gets back from the database the same reference records pertinent to the other recent loci it looked at. Maybe we can come up with 'safe' rules for regurgitating it from some cache, saving the bucket scan and bcf_unpack each time. The new query has to be contained within the reference record range, and there have to be no other records (in the same sample) overlapping that reference record.
Thread contention on this cache might become a problem, though.
When loci are sufficiently dense, most of the time the genotyper gets back from the database the same reference records pertinent to the other recent loci it looked at. Maybe we can come up with 'safe' rules for regurgitating it from some cache, saving the bucket scan and
bcf_unpackeach time. The new query has to be contained within the reference record range, and there have to be no other records (in the same sample) overlapping that reference record.Thread contention on this cache might become a problem, though.