Thanks for the explanation of the problem domain, Dr DB.; I appreciate it. On 02/16/2011 10:42 AM, David Boyes wrote: >> I, too, wished IBM would do more to highlight the z series >> capabilities in public forums, but Watson is a massively parallel >> system (2K+ cores, I believe), and I have my doubts as to whether >> or not a collection of z10s could be integrated together tightly >> enough to meet the software's requirements. I could be wrong, >> however. :-) > > It could be done, but this is a case where it would have been the > wrong choice. > > This is a vector-oriented SIMD problem (enormous numbers of simple > comparisons on bits of data to determine if it has a right answer or > not), and it's embarrassingly parallel in nature (very little shared > data between comparisons, few locks, few context switches required). > This is POWER's sweet spot, and forcing a Z into this role is > probably not a good idea. This is not the kind of problem the Z arch > is designed to solve. >
-- Dave Jones V/Soft Software www.vsoft-software.com Houston, TX 281.578.7544
