The industry is making good strides to prevent a repeat of the missteps in the 40G generation.
Cloud service providers are entering a mega-upgrade cycle driven by the increased demand for capacity and aging infrastructure in 2017. This cycle has triggered a guessing game for companies in the equipment supply chain from Ethernet switch to silicon and optical component manufacturers about the speed in which to invest. Will the dominant speed be 100G, 200G or 400G?
The stakes are high for companies up and down the supply chain. Those who bet correctly stand to win a pot of gold. When these companies consider their options, they are weighing a number of factors such as customer demand, price and availability of optics and switch system design.
Worldwide, the leading cloud providers are clearly split.
Through our interviews, we have concluded that three of the four largest Western cloud providers have a preference for 400G. The fourth, Google, appears to be favouring 200G.
At the same time, some OEMs in China appear to have an inclination for 200G. This implies China operators and cloud providers are interested in 200G.
We believe that 400G offers the following advantages over 200G:
- 400G optics achieve better economies of scale and steeper price erosion at high volume.
- Deployments fit on one rack unit, critical for achieving top-of-rack deployment for three of the top four cloud providers.
- A one-rack unit with 32 ports of 400G will be less expensive to build than a two-rack unit with 64 ports of 200G.
- Vendors such as Arista Networks in Ethernet switches, Broadcom in the silicon, and Oclaro and Luxtera in optical components are currently placing their bets on 400Gbps.
The data centre will go through at least two product cycles over the next five years. The first will ramp this year and will be based on 25G serdes, primarily using Broadcom’s Tomahawk chips. This will drive the data centre away from 40G deployments and toward 100G.
The second cycle will start in 2019. It will be based on 50G serdes, enabling 32 ports of 400G per one-rack unit.
Dell’Oro Group recognises that optical component vendors face the highest risk during these transitions. Transceivers are rarely flexible in the speeds they can support. Optical vendors must bet on one specific speed now—years in advance of product deployments—with just a few possible permutations.
Many equipment vendors lack funds to invest in both R&D and testing a variety of speeds. In addition, they want to be certain they are not investing in a technology that could have a similarly short lifecycle as past 40G products.
Dell’Oro Group views this situation differently from what we saw with the 100G transition. At that time, many optical component vendors invested in low-volume 100G Ethernet CFP/CFP2 only to miss the ramp of high-volume 100G Ethernet QSFP28. We believe optical component vendors have successfully built direct relationships with the large cloud providers and are now in a better position to assess high-volume market opportunities.
We forecast that manufacturer revenue for the Ethernet switches in the data centre market will exceed ₹95,192.77 crore ($14 billion) by 2021. During the next five years, Dell’Oro Group predicts revenues for speeds at or greater than 100G will be about ₹27,197.93 crore ($4 billion) in 2017, ₹47,596.38 crore ($7 billion) in 2019 and ₹67,994.83 crore ($10 billion) in 2021
Dell’Oro Group believes the industry is making good strides to increase collaboration between vendors and end users to prevent a repeat of the missteps in the 40G generation. As we scan the tea leaves, early indicators suggest that 400G will deal a winning hand.
Sameh Boujelbene is senior director of data centre and Ethernet switch market research at Dell’Oro Group.
First published by EE Times U.S.