OmniPath a threat? Paugh, says Mellanox, feel our gorgeous 200 Gbps

Data centres to get speed boost on existing cable


Mellanox reckons it can shrug off the threat posed by Intel's OmniPath technology, announcing silicon photonics-based devices running at 50 Gbps per QSFP lane for a total of 200 Gbps.

Arlon Martin, the company's senior director of product marketing, told The Register's networking desk that the transmitters and receivers it's developed are already capable of running at better than 50 Gbps.

The silicon photonics devices demonstrated at this week's Optical Fiber Conference in the US will, the company says, form the basis of a suite of new products: direct attach copper cables for 50 Gbps and 200 Gbps for in-rack applications and silicon photonics optical transceivers for 200 metre and 2,000 metre reaches.

The big pitch for release that “everything is backwards compatible”, Martin said.

The silicon photonics transceivers are supported by fibre and copper cables.

If 2,000 metres sounds excessive for data centre applications, there are to things to keep in mind. First, it's easy to rack up distance on a campus-scale bit-barn; and second, as Martin said, distance is frequently a shorthand for link budget, because cables go through patch panels and connectors.

“Often we use that reach number as shorthand for a 3.5 dB link, so it might be 500 to 1,000 metres plus connector loss.”

Martin told The Register the company expects 200 Gbps data centre connections to start taking off during 2017 and accelerate during 2018, although “there are companies that would take it today, if it were available”.

“We see servers moving to 50 Gbps connections, and if you have a 50 Gbps server, the 4:1 breakout the data centres user means they'll want the network to run at 200 Gbps – just like networks needed to run at 40 Gbps when we had 10 Gbps servers.”

Sticking with the QSFP form factor maintains the symmetry that already exists when 200 Gbps networks are running through the data centres.

The key to Mellanox's silicon photonics work is a Franz–Keldysh modulator that's “on the order of 40 microns long”, Martin said, while much of the industry is working on modulators “10 to 40 times larger”.

“We've driven the modulator to 50-60 GHz, so we know it's ready for 50 Gbps channels”, he added, and “it's the same modulator we already use for 4 x 25 Gbps … it's hard to tell the difference between a 100 Gbps and 200 Gbps chip.”

The other key announcement at OFC is Mellanox's interoperability demonstration with InnoLight Technology, trading photons at 1300 nm and 1550 nm.

The different wavelengths, Martin said, reflect technologies with different heritage.

“The reason 1550 nm is used in the telecommunications world is that they want devices to suit the fibre amplifiers and DWDM systems,” he said, “while 1310 nm is typical in shorter reach applications.

“We use 1550 nm because it works better with silicon photonics systems.” Others like InnoLight are using 1310 nm transceivers.

The interoperability demonstration showed that the transmitters can operate at their native wavelengths, because optical detectors are “very wideband, and don't really care about the wavelength”, he said. ®

Similar topics


Other stories you might like

Biting the hand that feeds IT © 1998–2021