Today Facebook announced the Open Compute Project which opens their innovative platform and data center specifications to the industry and showcases the advancements in their new Prineville data center. The data center has a remarkable design PUE somewhere below 1.07 according to Jay Park, Facebook’s Director of Data Center Design and Construction. This implies less than 9% of the energy consumed in the data center goes to needs other than the IT equipment.
What is even more remarkable, however, is the extent to which Facebook has achieved improvement in what we’ve called “Grid to Gate” efficiency. PUE only measures improvement of power distribution and cooling equipment in the data center. Once PUE reaches such low values other sources of inefficiency become more significant.
And this is where Facebook has excelled. With some willing support over the last 18 months from Intel engineers, Facebook has implemented several key design features, from carefully selected high efficiency power supplies and distribution equipment, to a novel “un-shadowed” platform design that optimizes thermal efficiency by enabling lower power, higher efficiency fans to meet system cooling requirements in a data center with higher ambient temperature.
Another outcome of the work with Facebook that benefited Intel was the development and maturation of new technologies. For example, the genesis of the reboot-on-LAN function (rebooting a server by sending a special packet through the network) was born from Intel’s work with Facebook.
Facebook has pushed their data center into a new regime of efficiency -- from beyond PUE to the regime of “Grid to Gate” efficiency. Congratulations to the engineering teams at both companies!
** Based on a 55% performance increase from Intel™ Xeon™ L5520 processor to the Intel Xeon X5650 processor using Facebook representative model benchmarks. Combined with Facebook’s overall 38% DC power reduction = 2.5 perf/W improvement = 60% W/perf improvement for a 60% reduction in power consumption per user. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. Configurations: Intel EPSD Willowbrook: 2 x L5520 2.26 GHz “Nehalem” CPU, 6 x 2 GB DDR3 DIMM 1066 MHz (12 GB), 1 x 7.2k SATA HDD; Intel EPSD Willowbrook: 2 x X5650 2.66 GHz “Westmere” CPU, 6 x 2 GB DDR3 DIMM 1333 MHz (12 GB), 1 x 7.2k SATA HDD. Test = Facebook “Dyno” Web Tier workload. Tests completed by Facebook “Labs” team with Intel support. For more information go to http://www.intel.com/performance