Cisco Unveils a Fully Integrated, AI-Optimized Networking Vision at Networking Field Day 39

I'm Tom Hollingsworth, event Lead for networking here at Tech Field Day, and here are my takeaways from Cisco's presentation in Networking Field day 39. Hi everyone. We just wrapped up networking field Day 39 and one of our presenters was Cisco.

They had a lot to say about AI networking, and it was one of the better presentations that we've seen in a long time because they really hit on a lot of the important things that you need to know about the way that AI networks are going to be created in the near future, and their technology has a lot of upside that you're gonna wanna know about. Here are my three big takeaways from the Cisco presentation at Networking Field, A 39. The first takeaway is that Cisco is delivering a comprehensive end-to-end solution that's focused on maximizing GPU utilization.

Their strategy is built on providing an integrated stack that extends from fundamental building blocks to the operational layer. The central goal of this strategy is to ensure that the expensive resources, the GPUs, are better utilized because network problems directly correlate with poor GP utilization. This end-to-end solution encompasses the entire infrastructure.

Cisco has their own special silicon, Cisco Silicon one that they're using to build on this, as well as supporting software like nx, OS and Sonic, as well as optics and an operating model to help you out. The foundational requirements for this mean that the network is no longer just a pipe. It has to be highly available lossless with a lot of bandwidth, resiliency, reliability, and security to help support massive demands all across the AI spectrum.

This can happen because Cisco offers reference architectures from enterprise all the way up to cloud to ensure that customers are using validated servers, networking components and storage, which simplifies deployment for first time AI users. My second big takeaway is that operational complexity is being reduced. This comes through prescriptive architectures and automated management platforms.

There's no denying that AI is a very complicated thing to design and operate, and Cisco is doing their part to reduce these challenges by offering simplified automated operational models, particularly for enterprises who may lack massive r and d budgets that you might find in your typical hyperscaler. When you look at solutions like Nexus dashboard, which provide a single pane of glass for provisioning, management, security, and analytics across all parts of the network, whether you're scaling up, scaling out, scaling across, working on your front end or your back Nexus dashboard has exactly what you need. Another key component is hyper fabric ai.

This is for customers that are seeking a hands-off approach. Hyper fabric AI can be delivered as a software as a service option that manages the AI fabric, which provides easy deployment, visibility and remediation, and it functions like a prescriptive model guiding users through those validated designs to make sure that what you're deploying is actually gonna work. Another key piece of hyper fabric AI is something they call run cabling, which automatically generates a complex rail optimized design map and provides onsite tasks assignment.

This is important because the system does not validate the connection until it is correctly cabled as per the plan. This helps eliminate mis cabling. That's a major source of performance degradation in the network, and it has been shown to reduce deployment time by up to 90% for some customers, while also validating that it's going to perform at a level that you come to expect.

My third big takeaway is that performance bottlenecks are eliminated with advanced congestion management and physical design. Cisco is employing technical innovations both in physical design and software, as well as other areas to ensure that non-disruptive operation is achieved by fighting things like delay in congestion. These are the leading causes of performance issues when it comes to these communication jobs.

As I mentioned before, rail optimized design ensures that the physical connectivity on the topology minimizes serialization, transmission, and propagation delays by connecting the nodes in a way that allows them to communicate with single hop forwarding. This means that your traffic isn't hitting the spine switches unnecessarily, and that everything is contained from point to point transmission. Another area that Cisco is using is dynamic load balancing.

This uses flow lits paired with NVIDIA's adaptive routing to solve congestion that arises from non-uniform link utilization. Dynamic load balancing detects groups of packets that belong to an RDMA operation and directs them to the least utilized and least congested link in real time. And when you're talking about thousands of links that are waiting on all of them to complete their tasks before they can return results, even the matter of microseconds can make a huge difference here.

All of this comes courtesy of the P four programmability of Silicon One asics, that results in perfectly balanced links and elimination of flow control activity benchmarks run by Cisco show that this combined approach delivers between 35 and 40% increases in bandwidth consumption compared to default ECMP approaches, which don't really work well in these highly congested environments. Through these big takeaways, I hope that you see that Cisco has put a lot of thought and engineering into the way that they're building AI networking for the future, and they understand the challenges that are inherent in bringing us into a new area of high speed, high throughput, latency sensitive networking. If you're looking to deploy AI networking in the future and you need to understand the best way to build it, Cisco has the reference architectures and the technology to make that happen.

Thank you for watching this episode of Tech Field a takeaways. If you enjoyed it, please make sure that you like, subscribe, and share your thoughts in the comments. Make sure that you're following Tech Field Day on X, Twitter, blue Sky and Mastodon for more updates.

And make sure you check out the Cisco presentation videos on the Tech Field Day website and on our YouTube channels.