May 1, 2014, Ethernet Technology Summit, Santa Clara, CA—John D'Ambrosia from Dell envisioned the paths leading to a TbE network. The efforts in defining the next-generation transport technologies is starting now.
Speed will always exist as a limiting factor in all electronics. For networks, the latest efforts are to get to 400GbE, even though many users are still on older technologies down to 10/100 base-T. The standards groups are defining 400GbE in the 802.3bs task force which started in March. Ratification is expected by 2017.
Until the 400G is ready, the industry still needs the two top speeds, 40/100GbE to address the changes in the core and server I/O data rates. Now networks need to have interoperability from 10G to 40G and 100G in a single system. Systems have 10GbE at the servers which may be feeding 1GbE to edge devices. The 40GbE is the datacenter backbone and the 100GbE powers the service providers. The fact that no one really knows the final configuration for 100 G further complicates the issues, whether it should be 4 X 25G or 10 X 10G.
The transport and modulation technologies to choose from are QSFP (Quad Small Form-factor Pluggable) and CFPfor optical modules. A multi-lane gearbox aligns the 10/40/100 data streams which feeds the forward error correction (FEC)circuitry. These streams the go to a modulator to feed the network links.
The users will pick cables for either speed or density, 4 X 10 for 40G or 1 x 40G. 400G can be 10 X 40G or 4 X 100G. The trend is to figure out how to build up the high end and manage the data flows. The tradeoffs between granularity and simplicity are complex. The drivers for higher bandwidth are in cloud-scale datacenters, Internet interchanges, wireless infrastructure, and video distribution. A major ecosystem issue is the need for back compatibility to the slower data rates.
For the 400GbE families, transport is likely to be up to 500m in parallel single-mode fibers. This solution will depend upon the costs and the ability to develop the rest of the 400G infrastructure. Uses will need to look at future apps and technologies to determine the best path going forward. The technologies will become more complex since latency and FEC efforts stack up.
The potential for electrical signaling with 25G lines would be possible with a 16 X 25G configuration, and might be upgradeable to 50G in the future. The proliferation of lines may also invite looks at non-serial transport. For example, an optical link would have lots of choices at 100G per wavelength. A serial protocol is not the only possibility.
In addition, 16 X 25G may be too expensive and not dense enough to fit into a backplane or front panel. For shorter reach, there is already a specification in process for 100Gbase-T, which would be much more space efficient. The defining parameter may be power.
The Ethernet industry is moving beyond 10G and will soon be above 25G. To get to 400G and eventually to 1TbE will require getting around the pesky laws of physics. The industry needs to perform lost of R&D to enable the transition to the next speed
