On Tue, Dec 20, 2011 at 09:28:05AM +0000, David Corking wrote: > Google engineers have said that a key design parameter for their > services is the cost of instructions in kilowatt-hours. So the problem
The costs of energy have been dominating the data centers for many years now. This is the reason for introducing higher temperatures, water cooling, ambient cooling, more energy-efficient nodes (ARM, Blue Gene) and so on. This will eventually lead to fully static designs, most likely spintronics which only burn power when switching. It's obviously the only way to achieve 3d integration, along with reversible computing. > may be less about idle cycles than about wasted cycles such as context > switches, inefficient algorithms and compiler optimization. > > Meanwhile, I suspect servers for interactive services are rarely more > than 90% idle. Clusters are computers, too. These are almost never idle, and don't context-switch (typically dedicated to scheduled batches). > So I would suggest that efficient scheduling and request run-time There's a tentative trend for allocating invididual dedicated nodes to particular tasks (instead, of, say threads), such as a sea of ARM cores on a mesh fabric. This seems to match cloud workloads quite well as well. The next years should be rather interesting. > limits are important economic and environmental problems to solve, but > the solutions are unlikely to eliminate many digital tech jobs, as it The first jobs to go are that of server monkeys. Lights out data centers do exist, and many developers are already using comfortably abstracted hardware instantiated, scaled and destroyed by APIs. This is dangerous, but also convenient. > is unlikely to benefit society by more than the equivalent of two > Moore's law doubling cycles. That is a huge prize, but not the end of Intel is down to 14 nm. Not many more where these came from. > the world as we know it. _______________________________________________ fonc mailing list fonc@vpri.org http://vpri.org/mailman/listinfo/fonc
