On 9/9/2010 9:49 AM, Kyle McDonald wrote: > > It never shipped, (well it did kinda, but not in it's full glory) but > the group I was in at Sun had developed HW to connect multiple E6K, > and E10K machines (up to 16 if I remember correctly) together, and > scale a single kernel instance across all of them. > Sorry, actually, I have the models confused. It was developed for multiple F6000/F6500, and I think it was planned to work in the F15K also. The HW did ship to some limited customers I beleive, but only as a HPC shared memory interconnect, (RSM might have been the product name) between multiple kernels. I'm pretty sure we had the single kernel scaling well in the lab though before the project was scrapped.
If anyone cares there was an earlier project to do a similar connection between 4 E6000's, but that was killed when Sun bought Cray, and acquired what became the E10K. After the project for the F series was killed, the same group started working on parts of the system (Millenium/Eagle) that would follow the F series. At one point there was talk of doing it again running 1 kernel on multiple boxes over IB, but that whole system was scrapped in favor of Niagra/Fujistu systems in the short term, and 'ROCK' in the long term... But but ROCK was also scrapped not too long before Oracle bought Sun. As others have said Sun was great as the engineering side of things. Not so great at deciding what to (or not to) engineer, and not the best at marketing the things they did engineer and build. -Kyle > There just weren't enough places interested in buying a machine that > large. > > -Kyle >
_______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
