[GSoC] 7th coding week

[Ketil Malde]

> Do you suppose the performance tradeoffs would be any different after the
> application is parallelized? (For example, some portion of the workflow
> that is more difficult to parallelize begins to dominate CPU usage -- or
> perhaps one data structure parallelizes better than another.) If so, would
> it make sense to first do the parallelization as you described in your
> project plan, and then return to profiling-guided optimizations?

I would say the biggest problem is the memory consumption.  Increasing
parallelism will likely only make it worse, and if memory can be kept
low, it's possible to exploit parallelims by running multiple alignments
in parallel (typically, you have a set of sequences to align that is
much larger than the number of available CPUs).

-k
-- 
If I haven't seen further, it is by standing in the footprints of giants