Newest Pindel code? You could split by smaller segments (say 10MB) per job with 4 cores and this will not affect interchr prediction.

If on exome data, I sometimes saw coverage spikes (several kx) in several narrow regions, this causes Pindel to slow down and have memory issue. You might use -J to exclude those regions. I have tried to solve this but have not found a smart way to handle huge coverage variation in the data.

kai

If you are interested in indels, I suggest trying BBMap. It allows alignment across long indels so you can detect them in a simple mapping+pileup rather than using an expensive realignment. It also allows an arbitrarily large number of indels or substitutions per reads, so has very high sensitivity to multiple events.

Note - while it can detect deletions of arbitrary length (depending on the maxindel flag), it cannot detect insertions more than ~50% of read length, since it only uses information from individual reads. So it's not a complete replacement for Pindel.