Bill Carter, Open Compute Project, Decathlete Server Board Standard, here. Lookout Magoo. To play this game I would expect that you would need to reevaluate and possibly retool your trading application code end-to-end about every 18 months. The AVX2 retool would deal with the majority of the restructuring I suspect, so the dev cost is mostly upfront unless tech really changes dramatically post Haswell and Broadwell. You would look for the Broker Dealer to deal with the System test, burn in, and OS tweaking risk and cost suitably mutualized. Have to look at the numbers but it is conceivable in steady state you could maintain a ~1.5 generation compute advantage over the the top performing Magoos.
The purpose of this document is to define a dual socket server board that is capable of deployment in scale-out data centers as well as in traditional data centers with 19” rack enclosures. Considerations are made in the specification for suitable two-socket server boards that were in production when the specification was released.
Open Rack, here.
The Open Rack is the first rack standard that’s designed for data centers, integrating the rack into the data center infrastructure, part of the Open Compute Project’s “grid to gates” philosophy, a holistic design process that considers the interdependence of everything from the power grid to the gates in the chips on each motherboard.
Sarti and Bratach, Open Compute Project, Deploying OCP Hardware in a Collocated Facility, here.
For more information about the Open Rack Triplet, read the following specifications: • Open Rack v0.6: http://opencompute.org/wp/wp-
content/uploads/2012/09/Open_Compute_Project_Open_Rack_v0.6.pdf • Battery Cabinet Hardware v1.0: http://www.opencompute.org/wp/wp-
content/uploads/2011/07/DataCenter-Battery-Cabinet-Specifications.pdf (if you choose to use the OCP Battery Backup Cabinet for backup power instead of facility power, in conjunction with the OCP Power Shelf)
• Power shelf v0.3: http://www.opencompute.org/wp/wp- content/uploads/2013/01/Open_Compute_Project_Power_Shelf_v0.3.pdf
Steven Levy, Wired, How Google Will Use High-Flying Balloons to Deliver Internet to the Hinterlands, here. Think I heard a quote of 1 million USD per month for London to NYC RF Balloons. I’d put a bunch of traders in several balloons for the crossing latency lock. The width of the transAtlantic lock window is so large you could launch total POS performing mil spec hardened silicon in the balloon for the trader. Alfred, please prepare the DynaPie RF Balloon, I’m ready to trade.
Attached to the bottom of each envelope is the 22-pound “payload.” It’s topped by a sheet of solar paneling the size of a basketball backboard. Beneath the solar sheet is a construct resembling a large camera tripod, whose legs are antennas that allow the balloons to transmit to their peers in a mesh network. And on the bottom of the structure is a metal-sided container resembling a deep fuse box, which contains the computers, electronics, GPS devices, and batteries to store the energy gathered by the solar panels (each about 10 times the size of a laptop battery). It also controls valves that go inside the balloon’s internal chambers, allowing the balloon to find the desired altitude to maintain its flight path. Dangling from the box is a cable ending in a piece of foam that looks like a slice of a kid’s swimming noodle; inside is a transponder that beams location to air-traffic controllers and other trackers.