Recently, Intel and facebook banded together to create the next-generation of data center rack technology. The prototype includes an innovative type of architecture called the photonic rack architecture. The design of this new prototype improved every aspect of data center racks including the design, cost, and reliability by implementing a disaggregated rack environment, Intel switch, and silicon photonics technology.
What are Silicon Photonics
Silicon photonics take advantage of light photons to move gigantic amounts of data at extremely high speeds over a small, thin optical fiber. Traditionally, to move data, electrical signals were used and sent over a copper cable. The prototype that Intel came up with can move data at up to 100 gigabites per second. This speed allows the components to work together, even when they are not in close proximity to one another.
New Options in Design
Intel has created a rack that separates components to their own server trays, one tray for atom CPUs, one tray for Xeon CPUs, and another for storage. This design is great because when a new generation of CPUs is on the market, the user can swap out the CPU tray instead of having to wait for an entire new server and motherboard design.
This design approach enables the independent upgrading of compute, network and storage subsystems. This independent upgrading ability will absolutely define the future of datacenter designs for the next ten years. Intels photonic rack allows for fewer cables with an increased bandwidth, power efficiency, and farther reaches than today’s copper based connections. These new technologies makes hardware much more flexible, and when coupled with the silicon photonics, enables interconnection without much concern over physical placement.
The term ‘rack disaggreagation’ simply refers to the separation of resources that exist in a rack, this includes storage, networking, comput and power distribution. The separation is in the form of discrete modules. The traditional arrangement of a data center rack would have a server with its own group of resources. When data center racks are disaggregated, resources are distributed and grouped by their types and upgraded on their own pacewithout affecting others.
Disaggregation not only increases the lifespan for each resource, it enables IT managers to be able to replace individual resources rather than the entire system. This modulation makes data centers much more flexible and serviceable which subsequently improves the total cost for an infrastructure rehaul investment. This arrangement also improves thermal efficiency because it allows for more optimal placements within a rack.
The optical interconnects today generally use a connector called MTP. The MTP connector was not optimized for data communication applications, it was designed in the 1980’s for telecommunications. Even though, at the time it was created, MTP utilized the state-of-the-art technology, it has not kept up todate. Many parts of the MTP connecter utilizes parts that are individually expensive and can be easily contaminated by dust.
New Connector Design
In the last 25 years, there have been significant changes in materials and manufacturing technology. Utilizing this technology, Intel, with the help of optical fiber and cable specialists, designed a brand new type of connector that uses modern technology and manufactoring technique. They have included a telesckiping lens to help prevent dust contamination, as well as used fewer parts with up to 64 fibers in a smaller form, all at a lower cost than MTP.
The new Intel prototype utilizes silicon photonics technology as well as distributed input/output using Intel’s ethernet switch silicon. This prototype also supports Xeon process as well as next-gerneration system-on-chip Atom processers. These innovations dovetail nicely with many other ongoing Open Comput projects. The SOC/memory module was created with the writing of CPU/memory ‘group hug’ module specifications that were proposed by Facebook. The exisitng OCP windmill board specification that supports the 2S Xeon processors, is planned to be modified to illustrate that the signal and power deliviery were modified to be able to interface with the OCP Open Rack v1.0 spcification for power dlivery through 12V bus bars, as well as for networking, to allow for interfacing with a tray-level mid-plane board that contains the switch mezzanine module.
Intel is also planning to contribute a desing that enables a photonic recptacle to the Open Compute Project, and plan to work with Corning and Facebook to standardize the design.
Other New Features
Intel has been highly involved and added several innovations to the Open Comput Project. Their innovations include new storage technologies, racks, and systems. Specifically, Intel has been working on finalizing the Decathlete board specification for a general-purpose, dual-CPU motherboard, large-memory-footprint for enterprise adoption.
What is the Open Compute Project?
The Open Compute Project is an initiative announced in 2011 by Facebook to openly share datacenter product design. This initiative began after Facebook redesigned their data center in Prineville, Oregon by Frank Frankovsky, the Open Compute Project leader. The design is still a long way from being used in data centers, but some aspects that were published have been successfully used in the Prineville center to help them increase energy efficiency.
The future of data center design is being created right before our eyes. If successful, the Open Comput Project will enable a rapid technological increase that will make data safer and more accessible than it has ever been before. With the new rack, connector and photonics, data will be much easier to store, use, and share. Collaboration is the key, which is why the Open Compute Project is so widely accepted by many diffreent companies across the globe. The future is coming, are you ready?