[Bioclusters] FPGA in bioinformatics clusters (again?)

[Kathleen]

> TIA:
> 
> One thing to consider when using FPGAs for bioinformatics or any 
> complex computational life science application is whether or not the 
> FPGA supports cross coupled communication.

... which is needed in which algorithms?



>From what I'm learning, cross communication is less likely an issue in
transactional-based problems.  Protein-folding would be a great example of a
non-transactional problem that is tightly coupled (needs cross
communication) and benefits from being spread across multiple nodes.  It
sounds like the re-programmable FPGA's might do cross-communication.  I'd be
curious to know at what cost and what does the overhead look like to submit
job, scatter job, compute job, and gather results.  We have proven linear to
super linear scaling on COTS, with negligeable overhead.  Node count for our
technology is unlimited.  Our scaling drops off when we run out of work. 




> I believe FPGAs do not, therefore, FPGAs will be limited in 
> scalability, performance and very expensive.

FPGA performance on various algorithms is already 1 to 2 orders of magnitude
(e.g. 10 to 100 times) faster than single CPUs.  The price points of modern
FPGA boards with modern algorithms is about 2-4x single node pricing.

As FPGAs improve (and they are), expect to achieve significantly more
performance.



Can a user intermix FPGAs with COTS-based technology or do they need to
scrap their whole investment?  What if one FPGA on a board fails?  How easy
is it to swap that out?  Do you swap out just the failed FPGA or the whole
board?  Who would swap that out?  What happens to the work allocated to the
failed FPGA, does it get dynamically re-directed? What is the max # of FPGAs
for a single board and does each board cross communicate?