In a talk entitled "Watts, faults, and other fascinating 'dirty' words computer architects can no longer afford to ignore," Google engineer Luiz Barroso explained the company's desire to achieve optimal data efficiency for its data centers. "Power/energy efficiency and fault-tolerance are central to the design of large-scale computing systems today," said Barroso. "And technology trends are likely to make them even more relevant in the future, increasingly affecting smaller-scale systems." Building a data center costs more than powering it for 10 years, so the goal is to build them to use as few watts as possible, since price is directly related to watts. In its research of its own data centers, Google found that warm temperature was not a significant factor in failures. In its efforts to reach optimal usage, Google has accepted that single-thread performance "crashes into a power wall," and that distributed shared memory systems can be crippled by "a single fault," Barroso said. He imagined a day when power companies gave away servers for free so long as the customer signed an energy contract. Google is currently working with partners, including Intel and AMD, "to create open standards for higher-efficiency power supply units," Barroso said. The applicability of multicore processors and growing parallelism is subject to the ability of programmers to develop efficient and concurrent programs. However, this could be easier for Google, since it is working with an extremely large amount of data. Along with fault-tolerant software, Google uses System Health Infrastructure to provide additional monitoring of servers. This technology may be open-sourced, but "some of this is infrastructure and we build it so intertwined with other software we have that it's hard to pull things apart," Barroso said.
