If you have not read the article on high performance math libraries in last month's 'Dr. Dobbs Journal', you can go to http://www.nag.com/local/whitepapers/index.asp and download the last paper in PDF (titled 'High Performance Math Libraries'). It can start you thinking. The article uses the OpenMP model of multi-threading to spawn multiple floating-point threads (for AMD Athlon64 computers) when a call is made to a BLAS subroutine. BLAS = Basic Linear Algebra Subroutine library. BLAS 1 = vector-vector; BLAS 2 = vector-matrix; BLAS 3 = matrix-matrix. The result is a bunch of lightweight threads dedicated to floating point operations that can (hopefully) fit within a cache and expand to simultaneously use every FPU available in a SMP computer (multicore and multiprocessor). All this is invisible to the scientist writing the program, who only needs to know basic programming. Even our current client would allow multi-thread processing of this type (I think - my personal opinion). But Rosetta began before SMP computers were widespread. It was a 500-MB Fortran program with static memory allocation when Rick Alther got hold of it. You really want to think about multi-threading right from the start when designing a program. Otherwise, why try to restrict yourself to BLAS calls? Until SMP became a reality, there was no reason for anybody to think in terms of the tricks of the trade in using multi-threading.

I assume (I have not been told) that the future Linux x86 client will handle multi-threading. Even so, new programs custom-designed for multi-threading will be more efficient than older programs that just grew.

Just my opinion,
Lawrence

I would check out the very generous Austin Energy Solar rebate program. I installed photovoltaic solar panels on my roof which gives me up to 3000 watts of extra power when the sun is burning down on us. This helps reduce the cost of running the A/C, helps power my array of computers, and runs my meter backwards if I am generating more than I can use. Furthermore, since the panels shield the roof somewhat from direct sunlight, that helps cool things too.

Did you know that if you covered about a 100 mile by 100 mile square of Texas with these solar panels, you would have enough electricity to run the whole United States, provided you also build a storage system to handle nights and dark days?

The current agent doesn't support multiple processors and nor does our version of Rosetta. Somewhere else in these forums/faqs, this has been discussed before, so sorry, I thought your post was a subtle hint for us to hurry up and support this. We are looking for ways to take advantage of multiple processors in future projects.

There do seem to be a handfull of people who have lots of machines contributing to our gird, but the bulk of the contribution is still from the many thousands of individual machines. So, the skew is nothing like 90 vs 10%. For the current project, please don't hold your breath. We need you around for the next projects too Thanks for your contribution!