When Shared==Local, disabling distributed bioseqs saves
a lot of time.
I tried just sticking a (disable_mpi_db = true) in the
part of the code that checks if Shared == Local, but
that didn't work.
With --disable_mpi_db 0:02.13elapsed
without 0:03.57elapsed
Logged In: YES
user_id=756995
Last I checked, --disable-mpi-db was incompatible with
translated search types. We could try disabling it by default
for blastn and blastp.
It does make sense to have the writer go directly to the shared
DB for its bioseqs rather than having the worker go to the
shared DB and then send the bioseq to the writer. disabling
the distributed bioseq lookup in this case should reduce
contention on the NFS server, unless the workers have cached
the bioseqs already.