Cisco is bringing high-speed Ethernet to the data center with four new switches it says will ultimately become the foundation of high-density, intent-based networks.
The new 400G Ethernet switches fall into two camps, two boxes in the Nexus 9000 family aimed at large-enterprise network customers and two in the 3400-S class targeted at high-bandwidth hyperscale users. These switches pave the way for Terabit Ethernet switching to become a mainstream technology, wrote Thomas Scheiberote, Cisco vice president of product management, Data Center Networking.
The Cisco announcement is the third 400G Ethernet switch pronouncement in the past few months – this summer Juniper made a wide-ranging commitment to the technology, and in October Arista joined the march with switches capable of supporting 400G speeds – both aimed at hyperscale cloud and data-center networks.
“The 400G timeline will be driven mostly by when the optics for the technology are available in scale and cost,” said Sameh Boujelbene, senior director at Dell’Oro. “We see a small ramp up in 2019 – though I would also expect to hear a lot more from Cisco, Arista and Juniper next year – but then 400G ports will more than triple in the 2020 timeframe.”
Cisco says 400G will let webscale customers, data-center and 5G-mobile operators create more powerful networks more cost-effectively while offering customers with four times the bandwidth and scale of current networks.
The bandwidth and scale needed for 400G is evident, Cisco says, because:
- There will be 20 billion to 30 billion online IoT devices in the world through this decade.
- Users will own more than 25 million 5G-capable devices by 2021 and more than 80 percent of all IP traffic will be bandwidth-hungry, latency-sensitive video content.
- By 2021 hyperscale data centers will grow from 338 in number at the end of 2016 to 628, representing 53 percent of all installed data-center servers by 2021.
Enterprise users may be most interested in the two new Nexus 9300GX boxes, which Cisco says expand its overarching Application Centric Infrastructure (ACI) and its intent-based networking data-center designs, which in real time can identify and react to changing network conditions.
ACI brings with it support for Cisco products such as the Tetration Analytics platform, which collects all manner of telemetry information from Cisco and non-Cisco devices on the network. The system then uses machine learning and behavior analysis to provide network administrators a greater understanding and control of their data-center resources.
For example, Cisco said Tetration gives customers a single tool to collect consistent security telemetry across the entire data center and analyze large volumes of data in real time. In a multi-cloud enterprise, Tetration can lock down tens of millions of whitelist policy entries across thousands of applications.
Another key component is Cisco’s Network Assurance Engine (NAE) software, which verifies intent-based network behavior and ensures intentions have been translated into policies and orchestrated throughout the network.
“The approach Cisco is taking with these switches – emphasizing software features over the speed of the box is really new for them, but also gives customers an idea of what Cisco really sees as valuable – telemetry gathering, automation, management, security all driven by software not hardware,” said Boujelbene.
As for the actual switches, the Nexus 9316D-GX is a 1RU spine switch with 16 400/100/40G ports and the Nexus 93600CD-GX is a 1RU leaf switch with 28 100/40G ports and 8 400/100G ports.
“For the most part, these switches will be positioned for sale to service providers pursuing telco cloud, including 5G scenarios, and for high-end enterprises. It will take a little longer for these 400GbE markets to develop, especially the enterprise, but it makes sense for Cisco to have an across-the-board 400GbE capability,” said Brad Casemore, research vice president at IDC. “The backward compatibility of Cisco’s full array of 400-GbE data-center switches affords customers flexibility as they move forward.”
It’s also important that Cisco said it will bring its BiDirectional (BiDi) optical-networking technology to 400GbE, giving customers the ability to re-use fiber cabling in the context of an upgrade to higher-speed interfaces, Casemore noted.
Cisco said the other Nexus switches, the 3400-S series, are aimed at hyperscale customers. The two models: The Nexus 3432D-S switch is a 1RU switch that offers 32 400G ports, and the 4RU Nexus 3408-S lets customers mix 100G/ 400G connections. Both feature a 12.8Tbps chip with a 70MB buffer.
Overall, there were some striking items in this announcement, notably the silicon choice, Casemore said.
“Cisco did not disclose the Ethernet merchant silicon that it is using in Nexus 3432D-S and Nexus 3408-S, but I ‘ve done some research, and I believe those switches are powered by Innovium’s TERALYNX 400GbE switch silicon,” he said.
All the attributes of the TERALYNX Ethernet silicon align with the specifications on the Cisco datasheet, Casemore said.
“This is notable because Arista has leveraged Broadcom’s Tomahawk 3 network silicon for its 400GbE switches. Cisco is the first major data-center networking OEM to adopt Innovium network silicon,” he said.
“It will be interesting to see whether this confers tangible competitive benefits for Cisco in real-world customer environments, but Cisco clearly is aiming these Nexus 3K models initially at hyperscale and webscale cloud customers, and they hope to differentiate on the basis of low latency and low power consumption,” Casemore said.
In response, Cisco said it normally does not talk about its component vendors.
Fiber optics are another interesting wrinkle, Casemore said.
All of the new Nexus switches support quad small-form-factor pluggable double-density (QSFP-DD) high-density optical connectors rather than another optical technology called Octal small-format pluggable (OSFP).
Cisco is wholeheartedly supporting QSFP-DD optical connectors, which are a little smaller than OSPF connectors and align better with Cisco’s backward-compatibility focus. Arista will support both QSFP-DD and OSPF, but it is opinionated on the superiority of the latter and believes it will prove a better long-term option for its customers. To this point, Juniper joins Cisco in the QSFP-DD camp, but optics preferences of customers and customer segments will be an interesting subplot, Casemore said.
Cisco said field trials with customers start in December 2018, and the new Nexus switches will be generally available in the first half of 2019.