Sunday 25 August 2013

Intel teases 1.6Tbps optical interconnect, asks us to forget about Light Peak

Intel teases 1.6Tbps optical interconnect, asks us to forget about Light Peak

  • By Poonam Das
Router cabling


Ahead of the Intel Developer Forum next month, Intel and Corning are teasing a new optical interconnect technology capable of 1.6 terabits per second. Dubbed MXC, the interconnect is designed to supercharge the interconnection of servers in data center environments, where current networking technologies are struggling to keep up with the massive growth of cloud computing. The tech could eventually percolate down to consumer devices, though — don’t forget that Intel’s Thunderbolt interface was originally an optical interconnect called Light Peak, but switched to copper to reduce costs. Intel would love to bring optical interconnects to the consumer market, it’s just a matter of doing it cost-effectively.
Intel’s MXC sounds like it’s a brand new system, consisting of a new MXC connector that replaces RJ45 (the socket used by the Ethernet ports on the back of your wireless router) and small form-factor pluggable transceivers (SFP) — and a new fiber technology called Corning ClearCurve LW. The abstract on the IDF website says that Intel has been working with Corning on MXC for two years, with Intel bringing its expertise in silicon photonics and Corning focusing on the new fiber. There is very little information about the connector/interface, other than it’s smaller than RJ45 and SFP, and that the connector can carry up to 1.6 terabits per second (Tbps).
To put this into perspective, the current top-end networking standards are 40GbE and 100GbE, which support 40Gbps and 100Gbps respectively. The IEEE is currently working on a standard for 400 Gigabit Ethernet (GbE), which is due to be ratified around 2017. From the IDF abstract, it sounds like MXC is all about the underlying link technology — the connector, the cabling, the transceiver — while something like 400GbE would actually manage the signaling and data transfer over that link.
Light Peak (Thunderbolt)
Light Peak was such an exciting concept. And then Apple killed it.
When it comes to fiber-optic networking, Intel isn’t the first company that springs to mind — usually it’s IBM, Cisco, or one of the supercomputer makers like Cray. Way back in 2010, though, we wrote a story about Intel’s silicon photonics R&D department creating chips capable of 50Gbps, with the potential to scale up to 1Tbps — and now, three years later, it seems Intel has produced a transceiver that’s capable of even faster operation. Perhaps more famously, the silicon photonics lab also produced Light Peak — an optical interconnect that started at 10Gbps per channel, but which could scale up to 100Gbps. For a variety of reasons, though, Light Peak never made it to market and instead we got a copper-wire version called Thunderbolt — and even with the revised version (Thunderbolt 2) coming to market later in 2013, it’s still stuck at 10Gbps.
Rather than discuss whether Apple killed Light Peak/Thunderbolt, though, and USB 3.0′s unsurprising utter domination of the 2012-2013 Peripheral Interconnect War, we will quickly steer the conversation back towards MXC. With more and more computation being moved to the cloud, and internet usage growing at a rapid clip, the links between data centers — and between servers inside each data center — are getting steadily more congested. The deployment of 40Gbps fiber channel links between servers and data centers has relieved some of the pressure, but faster links will be required in the future. Technologies like MXC will surely help, but there are competing proposals to reduce congestion, too, such asbouncing 60GHz wireless network signals off the ceiling between racks in a data center.

0 comments:

Post a Comment