Exascale computing may be far in our future, but the branding folks at Force10 Networks have it for you on a label today. Announced back in March, 2009, the Force10 ExaScale E-Series of core routers and switches boasts impressive performance specifications, now spiced with leading power efficiency:
Data center power and cooling costs are forecasted to rise significantly in the next several years. Ironically, positive device attributes, such as performance and port density, are driving these higher costs. Consequently, we anticipate data center managers will more closely examine how switches are architected to minimize energy consumption,” said Kevin Tolly, founder, The Tolly Group. “Our testing indicated that Force10 clearly recognizes this ongoing concern and has demonstrated a critically important capability to combine robust performance features with very low power consumption.”
While the ExaScale series sounds like an impressive piece of engineering for the cloud, it sounds to me like the marketing guys got a little carried away. Exascale is about scaling up to a million trillion sustained 64-bit, double-precision floating point calculations per second. That will likely require an interconnect that hasn’t even be invented yet. To keep this straight, maybe the “Scale Out” guys needs their own holy grail term like “HooptyCloud” or something. Just saying.
“Exascale is about scaling up to a million trillion sustained 64-bit, double-precision floating point calculations per second.”
Actually, it is not just DP FLOPS that is driving exascale computing – exaOPS will be as important as exaFLOPS and it explicity mentioned in the various initiatives. And exabytes (memory) will probably be an even bigger challenge.