High-speed Ethernet bargains help meet data center needs

High-speed Ethernet is quickly becoming the networking norm as customer data-center servers grow to handle a ton of traffic from new, smarter applications, IoT devices, video and more.

According to IDC in the first quarter of 2018 overall 100Gb Ethernet revenue increased 83.8% year-over-year to $742.5 million and port shipments grew 117.7% year-over-year in the first quarter of this year. Researchers at Dell’Oro Group said they expect to see somewhere near 12 million 100G Ethernet ports ship this year compared to about 1 million 100G Ethernet ports shipped in 2016.

The need for increased data-center speeds is being driven by many things – the tremendous growth of hyperscale networks from players like Google, Amazon and Facebook, but also by the price/performance of 100G products, said Sameh Boujelbene, senior director at Dell’Oro Group.

A recent study from PwC went a bit further in explaining how and why the network might need more speed. “Workloads are becoming less monolithic as companies branch away from the traditional enterprise data center. They are becoming more distributed, more mobile, and more like the workloads typically associated with hyperscale environments,” PwC wrote.  “…Almost all major workloads [will be] shifting from on-premises to public cloud in the next 1-3 years. Applications will be more tied to the network, and the network will become more critical given the distribution/dynamism of the workloads.”

The requirement for more high-speed ports and more data being driven from the dense edges of the network is driving the upgrade of the backbone, said Roland Acra, senior vice president and general manager of Cisco’s Data Center Business Group.

“It’s driven largely by the evolution of the NIC – the network interface cards on the servers. Quite a few of the servers were largely 1G to 10G attached. Now, we’re beginning to see the pretty significant change on the servers that are attached to the top‑of‑rack switch and going up from 10G to 25 or 50, which is now making the uplinks from the top‑of‑rack, wanting to be higher density, 100G,” Acra stated. 

Cisco also wrote on its blogsite: While demand drives the need, it is the compelling economics of 25GE and a core layer with 100GE that will drive deployment.

Moving from 10G to 25G to 100G Ethernet

“Optics prices have dramatically dropped, with 25G at essentially the same price as 10G and 100G at the price of 40G. Using four lanes of 25GE, 100GE backbone platforms require lesser cabling and reduced space requirements and cost savings. Backward compatibility provides additional options to ease the transition while extending the value of your current assets,” Cisco said.

There are other forces at play, as well.

“Cloud and software-defined architectures are shaking up the Ethernet switch and router markets,” wrote Petr Jirovsky, research manager, for IDC’s Worldwide Networking Trackers. “Continued price erosion and increasing divergence between buying preferences of cloud and communications service providers and enterprises creates a challenging environment for vendors, but also opportunity for end users.”

Cisco, Juniper, Arista, HPE, and Huawei target data center

Cisco, Juniper, Arista, HPE, and Huawei are just a few of the vendors that have been aggressively pursuing the opportunities in higher speed as well as traditional speed Ethernet markets. Juniper most recently rolled out its EX4650 high-density 25/100 Gbps switch which supports 48 100G Ethernet ports, or 48 ports at 25G and eight 100G uplinks.

But the drive for 100G is only the start for higher-speed Ethernet. A Dell’Oro report earlier this year said 400G is forecast to comprise 20% of data-center switching revenue by 2020. Higher speeds – 100G, 200G, 400G  and 800G – are forecast to drive significant growth over the next five years.

“In December 2017, Broadcom announced its Tomahawk 3 chip based on 56G SerDes, joining both Innovium and Nephos,” said Boujelbene. “We expect merchant silicon based on 56G SerDes to drive growth in 200 Gbps and 400 Gbps shipments, with 400 Gbps comprising majority of the volume. By 2020 to 2021, we expect another speed upgrade cycle driven by 112G SerDes, which will drive 800 Gbps port shipments – plus another wave of 200 Gbps and 400 Gbps shipments.”

Cisco’s Acra said 400G is still two or three years away or at least until the 400G plummets enough to be attractive. “Still 400G would very much be a big cloud and service provider concern at this point. The enterprise hasn’t made 400G links a priority.”

Juniper all in on 400G Ethernet

Juniper, however, looks like it wants to make 400G Ethernet speeds a reality much sooner than that. This week it announced a wide-ranging roadmap on how it plans to transition its wide-area network, data center and enterprise portfolio to 400G. The company said it expects to take advantage of the recently upgraded silicon in its switches – ExpressPlus silicon for its PTX line, Penta for its MX series and Q5 for its QFX boxes.

Juniper said its roadmap will help cloud, enterprise and service-provider customers “recalibrate network economics as they transition to 400GbE to lower cost-per-bit while keeping up with increasing bandwidth demands from emerging 5G, augmented and virtual reality, cloud, and 4K video production and distribution.”

The company said it will integrate 400GbE technology across its PTX, QFX data-center and MX routing families for a wide range of use cases, including backbone, peering, data-center interconnect, scale-out metro core, telco-cloud services and hyperscale data center IP fabrics, all of which demand high bandwidth, low latency and security.

“2018 marks the start of the commercial 400G market, with volumes ramping up in 2019 as 400G trials across WDM, service provider routing, and data center switching applications convert into production deployments,” wrote Matthias Machowinski, senior research director at research firm IHS Markit, in a statement for the Juniper announcement. “We expect $10 billion will be spent on 400G technologies over the next five years.”