The Ethernet Alliance recently released its 2015 roadmap, which outlines Ethernet’s response to the ongoing desire for more bandwidth in data centers by adding new speeds for both in-rack and backbone connections. The roadmap calls for some eye-popping speeds in the future, but also charts out a plan for the low end of the market, representing an unprecedented level of activity for Ethernet.
The first new Ethernet speed is almost upon us. 25 GbE is designed to replace 10 GbE and 100 GbE will fit into the 40 GbE slot (both using quad-lane connections), basically by speeding up the individual link. The first 25 GbE gear should hit late in 2015 with production volume in 2016.
25GbE will help the demand created by virtual systems and the huge data volume of the Internet of Things, but clearly we’ll be back to network bandwidth under-provisioning in just a couple of years, so the roadmap foresees another doubling of speed, to 50 GbE and 200 GbE in the 2018 to 2020 timeframe.
Even these speeds aren’t going to be enough, based on a broad look at the industry in a 2012 study by an an IEEE committee. Consequently, the roadmap has plans for a 400 GbE link using 50 GbE technology in an eight-lane configuration. Beyond 400 GbE, dates start to get fuzzy -- “sometime after 2020” -- but doubling to 800 GbE will occur when true 100 GbE links are available. This technology will allow larger lane bundles, too, giving us a whopping 1 TbE or 1.6 TbE link with 10 or 16 lanes.
Of course, some of these high-performance levels will come at appropriately high prices, and within the basic roadmap we see a move to lower connection cost over time. The advent of single-lane 50 GbE will obsolete quad-lane 40 GbE, while the 2018 expected date seems to have made a 25 GbE dual-lane product unnecessary.
All of these products are aiming at the primarily fibre-based SFP and QSFP market, for primarily backbone or longer-haul connections. The venerable low-cost copper connection, which constitutes by far the bulk of connections, from servers to the top-of-rack switch, is still going to be available in higher speeds at 25GBASE-T and 40GBASE-T, although using upgraded Cat 8 cabling. These links will handle 30 meters of connection, which is more than enough for in-rack connectivity.
The Ethernet roadmap also addresses the low end of the market. Driven by a need to upgrade wireless hub connections without expensive rewiring, we are going to see 2.5 GBE and 5 GbE connections using Cat 5e out to 100m. This will support higher speed wireless hubs through to the end of the decade.
It’s been noted that the IEEE is planning more Ethernet standards in the next five years than its released in the last 40. This reflects the ingenuity and agility of the teams designing the core technologies involved, and presages further gains in the 2020s. There is even realistic talk of 10 TbE towards the end of the next decade.
Clearly, this rapid Ethernet evolution is needs-driven. The IO performance of flash-based storage, coupled with the horsepower of GPU and high-core-count CPUs is creating the need for a major boost in rates. The key question is affordability. New cabling is just the start of the upgrade cost, and it makes more sense to make an upgrade part of a new installation. This begs the question of when motherboard chipsets will handle the new Ethernet speeds; Intel, AMD and ARM will have to deliver their own roadmaps to clarify that.
So many new standards could lead to quite a bit of confusion in the next few years. There are Ethernet subcategories which aren't road-mapped by the IEEE, including RDMA, converged Ethernet solutions and accelerators for protocols such as iSCSI and FCoE. These make the roadmap blurry, so that elements overlap each other; creation of a clear plan for a data center will be challenging indeed.
One other impact of the intense emphasis on Ethernet is that other connection schemes will find their existence hard to justify. Fibre Channel and InfiniBand may be casualties of the rapid change, while FCoE could find itself in an inescapable backwater. With Ethernet drives now available, even SAS and SATA may fall by the wayside in favor of a single, unified approach.
Network administration has never had so many opportunities to solve problems and so many challenges in identifying the correct path forward. On the whole, this is a very positive trend for the IT industry, but not a free lunch!