Ilona: Hello and welcome to the first episode of Aviz Networks podcasts series, ‘Networking 3.0 Unpacked’. I am your host, Ilona Gabinsky, and with me today is Vishal Shukla, the Co-Founder and CEO of Aviz Networks. Vishal, welcome to the show.

Vishal: Thank you, Ilona, for hosting me on your podcast series.

Ilona: Yes Sure. Recently, Aviz announced the availability of the Networking 3.0 stack with the GenAI Network Copilot. For our listeners who might not be familiar, could you explain Networking 3.0?

Vishal: Yes. Certainly. So, to explain networking 3.0, I will explain it, analogous to, infrastructure 1.0, 2.0, and 3.0. And with that, I will explain how networking 1.0, 2.0, and now how it is in 3.0. So think of
infrastructure 1.0. For any infrastructure for that matter, you’ll have servers, storage, your mobile phones, or anything which is sold as a hardware plus a software combination. So 1.0 is all about innovation, a packed solution, which comes as a hardware plus a software and, you know, the pieces of application which are required to manage it packaged as 1. And that’s more about the hardware economy. Customer pays for buying the hardware, and they get a packed solution. That’s infrastructure 1.0. Analogous to that is networking 1.0 where customers used to buy the switches and and then the routers which came packed everything, and you just plug it in and everything works.

Right? So that’s your networking 1.0. Networking 2.0, like any other infrastructure 2.0, comes with some standardization on the hardware where your ASIC or the CPU starts to get commoditised, and then the standardization starts to come in the software layer, which gives birth to a lot of application driven economy. Now if you extrapolate it towards networking, networking 2.0 is based on the hardware and is based on the merchant silicon a 6. So kind of the switches got standardized.

Then you have the operating system and the number of protocols which makes the networking 2.0 based fabric standardized. This was essentially driven by the hyperscalers, the Cloud guys who essentially standardize the protocols, standardize on how they manage the networks using infrastructure as a code and civil Grafana based dashboards and everything. Essentially, some standardization on the hardware as well as software. That is 2.0. Now 3.0 is all about the data and the AI attached with it.

Now if you look at the infrastructure, it does not matter where your infrastructure is or what application that infrastructure is running. All that matters is that data coming out of it, how you are using that to put it in AI systems to effectively manage that infrastructure, first of all, and then making it future proof for various applications, which are yet to be discovered. Right? So something similar has to happen in networking 3.0, wherein it’s actually already started with Sonic coming in from the open source, more standardization which is happening. And because
of that standardization, the data lakes could be made for, multi vendor, ecosystem. And we are making and bringing networking towards that era of 3.0, which is all about data and AI, not only from the management point of view, but also finding the use cases for the future. So I hope you got this 1.0 hardware, economy, 2.0 application economy, and 3.0, data and AI economy.

Ilona: As I understand, the Networking 1.0 is all about hardware with comes set of protocols. Networking 2.0 is a commodity hardware with standard network operating system and template driven network management. And Networking 3.0 it’s all about data and AI.

Vishal: Thank you, Ilona, for summarizing it so well. Yes. Precisely. That’s what it is.

Ilona: But why the customers would care? What’s in it for them?

Well, to be very precise, we have summarized this in 3c’s, choices, control, and cost effectiveness. So if you start doing the standardization of any sort, right, that essentially enables a lot of choices for the customer.

Now think of it in this case is there are choices for the hardware and as well as choices for the network operating system. Now if you standardize things on the data lake and standardize the operations of that using the data which is coming from multiple networks, essentially, you have now a lot of choices, which means a lot of choices from switches, lot of choice choices from ASICs, and lot of choices from network operating system. So a lot of choices. Now with the choices comes the control. Now customers can actually control their network the way they want to control it.

And control is very important. You see what happened in, COVID era. There was a huge push on digitization, but at the same point of time, there was not enough supply for the switches. And actually who was in control? The vendors were in control.

Customers were not in control. So with the choices coming in, the customer gets into the control, and they can actually decide what is right for their data center and for their infrastructure edge or wherever they are putting the networks. Now with the choices and control comes the 3rd natural next step, which is the cost effectiveness. So, of course, once you have a lot of choices, a lot of control, then, of course, you can go up to 30-40% of savings. And we have done this thing for a lot of customers, so we know how to save, by providing the choices and control to the customers 30 to 40%
savings in their infrastructure. So summarized, three c’s.

Ilona: Yeah. I really like the 3 c value propositions, choices, control, and cost savings. It’s easy to remember. So now, can you elaborate a little bit on the role of AI in Networking 3.0, and how does it really fit with Network Copilot?

Vishal: Yes. Certainly. So, I see it in 2 dimensions. The first dimension is making the networking ready for AI workload, which essentially also known as AI fabrics. Having your networking stack, which can be
multi-vendor, so that you are not dependent on a single vendor. But still, it can help you make the AI fabric, which means understanding of end to end RDMA over converged Ethernet, PFC, ECN, and the right quality of service, and, the the buffering in the ASIC. Right? So essentially, that’s the first, use of AI and networking. So you need to make a network which can host your AI applications. The second one is essentially using the AI to efficiently manage your network, and that’s where the LLM driven scope pilot comes into the picture.

And they can help you too, in pretty much all the places. Now you can, you know, you can say that you have automation in place, but automation is not enough, if you actually compare what LLMs can do. You have 7,000,000,000 parameters which actually LLM can take care of if you talk about Mistral.

How much automation can take care of that?

Use cases like upgrade upgrade compliance, you know, checking your compliance metrics, and then, predicting your trans receivers and daily troubleshooting and everything.

Yes. It can be done by automation, but can it be future proof with those automation? No. So really, I mean, AI is impacting the networks in 2 different ways. First is actually and networks are being designed for the AI fabrics, for the AI workload. And the second one is AI being used to manage those networks. So customers should think of AI in 2 different dimensions.

Ilona: Yeah. This is definitely very interesting, and I’m sure it’s also very interesting for our customers. What recommendations do you have for getting started?

Vishal: Well, so, you know, I’ll again talk about 2 dimensions. 1 is the bottom of the stack. So if you are looking for upgrading your networks to the higher and better feeds speeds and feeds, so that’s where you can start with a multi vendor networks, introducing Sonic, and learn that how you can have up to 40% of saving out of that 3 c’s, you know, while increasing the choices and the control. So that’s one way of doing it. And trust me, I mean, we see customers from all different verticals, so it does not matter what kind of your network it is.

It sonic applies everywhere and, you know, this basically, you can talk to us, figure out that, your protocol supports it or not. So that’s the bottom set. Now talk about the top down. Top down means that you insert the generative AI based Copilot in your existing network. You don’t have to have Sonic or anything like that.

Just to start with AI on top, build the use cases, do the TCO saving and the return of investments using that AI, and then slowly go down the stack to change or to insert more choices in your network, which eventually will take you towards the cost savings, and up to 30 to 40 or 50%, even in some cases if you have the application. So essentially, you know, where to start? Start with Aviz networks, and we’ll suggest that, you know, you should start with a bottoms up approach or a top down approach. Both approaches are viable.

Ilona: Vishal, thank you for sharing your insights and the exciting developments at Aviz Networks. It’s clear that Networking 3.0 has the potential to transform the industry. We look forward to seeing how it unfolds. Thank you for being here.

Vishal: Thank you, Ilona, for, having me here. Thank you very much.

Ilona: Thank you for listening to our podcast. Stay tuned for our next episode where we’ll talk about AI fabric data driven network applications for pilots and so much.