Open Rack: Breaking the Mould

July 4, 2012

Facebook is working on a new, open standard for server rack design that integrates the rack into the data center infrastructure. Open Rack is part of the Open Compute Project’s “grid to gates” philosophy, a foundational technology that aims to diminish or eliminate the traditional differentiation at the mechanical rack and power distribution layer. Open Rack is an open design that focuses on enabling standard mechanical form factors, modular power supplies, improved cable management, efficient thermal design, and flexible serviceability allowing fluid component upgrades and independent refresh cycles.

The Open Rack project addresses the following issues:

  1. Different mechanical form factors require different implementation schemes and force vendor lock-in without associated benefit, as typified by blade servers.
  2. Redundant power distribution integrated into and specific to each server.
  3. Outdated cable management with proprietary interconnects.
  4. Impractical compute density, restricted by thermal design.
  5. Inefficient serviceability.
  6. In flexible installation processes.
  7. Inefficient component upgrade cycles.

The biggest challenge is the modification width of the server itself, expanding from the 19-inch width rack that has long been the standard slot for servers, to a new 21-inch format that will allow for improved thermal management, as well as better connections for power and cabling. Frank Frankovsky, director of hardware design and supply chain at Facebook and a co-founder of the Open Compute Project, justified the decision saying it was time to replace the veteran 19-inch rack with a bigger rack that would do a better job of packing compute, storage, and networking gear in data centers and provide better airflow over components. By sticking with the current 19-inch racks and their limitations, “we all end up with racks gone bad,” said Frankovsky, He explained that this dimension for racking and stacking gear was first used for relay switches for the railroad industry in the prior century and was subsequently adopted when telephone switching went from human-based to electronic when machines started getting racked up in volume in the late 1980s.

Although 19-inch racks were originally designed as mounting systems for railroad signaling relays, their format and dimensions remained basically unmodified after they were recycled for completely different industries. Besides being widely used for computer server equipment, 19-inch racks are also often seen housing industrial equipment, control and automation hardware, and in the entertainment industry, where they are used for  professional audio and video equipment, including amplifiers, effects units, interfaces, headphone amplifiers, and even small scale audio mixers.

Typically, a piece of equipment being installed has a front panel height 1?32 inch less than the allotted number of Us. A rack unit or U is a unit of measure used to describe the height of equipment intended for mounting in a rack (1.75 inches high). Thus, a 1U rackmount computer is not 1.75 inches tall but 1.719 inches. This gap allows a bit of room above and below an installed piece of equipment so it may be removed without binding on the adjacent equipment.

Computer servers designed for rack-mounting can include a number of extra features to make the server easy to use in the rack, such as sliding rails that can lock in various extended positions to prevent the equipment from moving when extended out from the rack for service and locking pins on the sides that just drop into slots on the extended rail assembly, in a manner similar to a removable kitchen drawer.

The Open Rack improves the configuration in several areas. By dividing each column in the rack into one or more power zones it allows to have an equipment bay housing all related components for the compute, storage, or other components and a power shelf, which powers the compute components in the equipment bay. The power shelf includes backup power capability using an external Open Compute Project Battery Cabinet, or it comprises a power shelf and Battery Backup Unit that embeds batteries. If there are multiple power zones, they are stacked one above another.

Each column has space for up to three Ethernet switches. In the current version, the switches are installed either at the top of the rack, above the topmost power zone or within its own equipment bay, typically above the power shelf, so the switch may be powered at 12V from its own power zone. Users may arrange the switches and equipment in other configurations as needed. Each equipment bay in a power zone can be set up in various configurations and can accommodate different numbers and height of equipment chassis (such a s individual server or storage chassis).

Motherboards and other Open Compute hardware are hot-swappable within the individual equipment bay . Variable configurations are achieved because the rack enclosure is designed to accommodate various equipment mounting configu rations. The rack has no side walls; rather, shelves in the equipment bay are locked in place to the vertical support posts. As long as the design follows Open Compute Project standards, the chassis (whether server chassis, storage chassis, or other) will fit.

Meanwhile other big players are also working on several hardware and software improvements for their data centers, although mostly focusing on achieving the optimal range of temperature and humidity for higher efficiency at lower costs, both financial and environmental. Microsoft is building a new chiller-less data center in Ireland and Google is currently operating one in Belgium. But these innovations are driven by a completely different approach. While Microsoft is using Direct eXpansion (DX) cooling which is similar to traditional air conditioning, Google has opted for a software system called Spanner that automatically moves and replicates loads between its mega data centers when the outside temperature gets too high.