Cisco Figures Out How To Sell Switches To Hyperscalers And Enterprises Alike

Selling hardware into the modern data center is no easy feat, especially when the needs of hyperscalers and enterprises have long diverged.

Hyperscalers and traditional enterprises have some overlap in requirements, but rely more heavily on some aspects of a system than others, compared to enterprises. Organizations want some bandwidth, but value longevity and consistency with previous systems. Hyperscalers must always lower bandwidth costs because their bandwidth requirements are always steeply increasing, and they must also lower the power consumption for each bit sent over the network. There are fewer hyperscalers, but they buy just as much equipment as the heavily based companies.

“Hyperscalers want to build ultra-dense, ultra-scale AI and ML networks,” said Gurudatt Shenoy, vice president of product management for the mass-scale infrastructure routing group at Cisco Systems The next platform. “Then the rest of the base is pretty much the full breadth of data centers. There are corporations, telcos, media providers, and they’re setting up data centers for all these different applications. Streaming increasingly going into richer formats, IoT, Edge, Compute, all of that. In short, the portfolio we’re bringing with us aims to cover this broad area of ​​our data center builds.”

For both hyperscalers and enterprises, the need for greater energy efficiency is at the forefront. Hyperscalers and cloud providers — think Google, Facebook, Microsoft, and Amazon to name a few — typically start planning their data centers with the power budget they have available, and then work from there to figure out how much to put in single device should consume. says Schenoy.

At the heart of what Cisco does with its data center networking equipment is its three-year-old Silicon One programmable converged switch and router ASICs that can be optimized for specific workloads. Cisco has deployed the Silicon One chips in all of its Catalyst and Nexus families of switches and routers, and is now leveraging its capabilities for switches unveiled at this week’s OCP Global Summit.

Cisco introduced two high-end 800Gb/s routers, the Nexus 9232E and the Cisco 8111. Both devices are housed in a 1U package and are powered by the 7-nanometer silicon One G100 chip. They can support 32 ports of 800Gb/s and can be broken down to 64 ports of 400Gb/s or 256 ports of 100Gb/s, providing even higher radix in tighter spaces by using breakout cables. The Silicon One G100 has an aggregate bandwidth of up to 25.6 Tb/s.

We detailed the G100 switch ASIC back in March 2021, explaining how Cisco has a modular architecture that covers both switching and routing – something the hyperscalers and cloud builders have insisted on, as Cisco is among them customers has re-entered competition with Broadcom. Incidentally, even Broadcom has different Ethernet ASICs for different tasks, although the hyperscalers and cloud builders may eventually push them to converge as well. Broadcom has “Trident” ASICs for enterprises that need maximum protocol support, “Tomahawk” for hyperscalers that focus on bandwidth and cost and need less protocol coverage, and “Jericho” with deep packet buffers targeting routing and some switching.

Here are the new 800Gbps Nexus switches from Cisco:

According to Cisco, these new switches can offer a 77 percent reduction in power consumption and an 83 percent reduction in footprint compared to alternatives.

“It’s a singularity system,” Shenoy said. “What is different about our 25.6T versus 25.6T from others that have announced is the 100Gbps SerDes. We are the first to ship 100Gbps SerDes on a 25.6T chip and offer native 100Gbps SerDes. That’s a differentiator, and once we start talking about 100Gbps SerDes, it means a lot of good things. It means you have a lot more energy efficiency. That means you can use 800Gbps optics as the standards emerge.”

The Silicon One G100 also comes with a wide range of features aimed at the AI ​​and machine learning workloads that are critical for hyperscalers and cloud providers.

Cisco also tends to offer both hyperscalers and enterprises systems that are optimized for their needs. Both have identical hardware, but the Cisco 8111 is a disaggregated system aimed at organizations that want to run third-party network operating systems. The Nexus 9232E is a highly integrated switch that comes with Cisco’s Nexus OS and its Application Centric Architecture (ACI) management stack.

“The Cisco 8111 is for a very small group of customers. It’s really something for hyperscalers,” Shenoy said. “It’s companies like Microsoft that want to put SONiC on it, or Facebook that want to put FBOSS on it. It’s a fraction of the customers. They have very high volume and deep pockets, but they are small in number and care about disaggregation. So we build this portfolio according to certain standards. Then the vast majority of the market – enterprise, telco, everyone else – wants a fully integrated solution. This is how we look at everything we bring to market. But again, the basics are the same: as differentiators, there are 100Gbps SerDes and efficient and sustainable platforms, but with these two different models.”

Less than ten years ago when we started The next platform, there was a rush by network vendors to develop disaggregated, open devices capable of running a range of network operating systems, and a very large number of proprietary NOS, as well as some new ones, were open source. The idea was to destroy the network appliance as the server appliance and storage appliance had been decades earlier. This disaggregated networking gives organizations more flexibility to run switches that best fit their environments and workloads—something the hyperscalers and cloud builders were demanding. Nevertheless, many companies have opted for it Not spend the money or resources to develop, operate, or manage disaggregated systems, according to Thomas Scheibe, Cisco’s vice president of product management for cloud networking.

“That seems to have completely shifted to enterprise interest in getting hybrid cloud deployments off the ground,” says Scheibe The next platform. “How do we connect instances from cloud providers to data centers that these enterprise customers operate? The focus has really shifted to the operational tools for the majority of these customers and how to do truly software-defined networking in terms of putting things together, basically truly hybrid cloud on-ramps and off-ramps. That seems to be of much more interest to them when it comes to what value they are trying to unlock in the future. It really depends on what type of customer you are.”

At the same time, Cisco introduced two high-density optical QSFP-DD800 form factor transceivers, which the company says double the bandwidth of previous offerings and allow enterprises to leverage the bandwidth on new 800 Gb/s platforms.

They also offer high-density breakouts to 400 Gbps and 100 Gbps interfaces and are backward compatible with QSFP transceivers. Shenoy said there is currently no standard for a single-flow 800Gb/s transceiver, but they are being developed, adding: “We are actively working with the standardization bodies there and as soon as it emerges we will support this.” , to.”

Market research firm Communications Researchers expects the market for 800 Gb/s and faster transceivers to reach global sales of $245 million by 2025 and grow more than 10-fold to $2.5 billion four years later. powered by workloads like video, 5G and more IoT applications.

The Cisco transceivers, which will be available in the first quarter of 2023, will also connect single-mode fiber optic links in the data center up to 2 kilometers.