Sramana Mitra: Help me highlight whatever other innovations and thought leadership you’re bringing to this venture. I don’t know your innovation portfolio. Give me some guidance on where else we should take this conversation.
Prem Jain: I can give my perspective, and I think Soni can add also. When we started looking at it at the beginning, we had 23 years of working with Cisco. We had designed many different networks.
When we were looking at this particular architecture, we said, “If we had to do this ourselves, we’ll take a solution approach.” What innovation can be done in hardware? We always look at the data plane, control plane, and management plane.
There was another design before Annapurna. They looked at one of our designs which we did as part of the NOAH systems when we were in Cisco. It’s how to make a distributed service card.
In that particular fashion, what are the particular innovations we can bring in? One is scale. Cloud needs to scale very well. Second is, it’s a distributed architecture. We need to make sure that it scales and also provides services where the server is sitting. This card was on the server.
Now whether you are in a virtual machine or container, we can provide those services consistently to all those use cases running on top of this infrastructure. And, we also wanted to make sure that it’s easily manageable. We provide the lifecycle management and make everything based on the policies.
A lot of people provide the SKD. A lot of people provide the chips. These vendors then try to figure out how to use it. It takes them a couple of years. We have accelerated that process by providing a stack which works very well in these environments.
On top of it, we provide the RESTful API so that they can integrate if they have their own controller or they want to use our controllers. We make sure that these software structures are modular enough so we can use the same technologies in multiple instances.
Innovation comes from one thing that we do differently. I’m pretty sure you’re familiar with the language P4. We looked at that. That language is good for doing the data processing part of it. We extended that part of the language. The extensions are already provided.
We customized those to our use case. That is differentiated value. That differentiates us in terms of performance, latency, and jitter. It’s totally programmable. Nobody has the technology that we have. It can be useful in multiple environments because we can implement a bunch of protocols.
If you’re doing it on the ARM processors, you will not get the same performance as you get on the Intel Core. We have broken that barrier. Control plane and management plane go into the ARM processor. The data plane is done in the P4 engine.
The other thing we’ve done is use high-speed memory. We can do networking services, encryption, compression, and other services.
Soni Jiandani: Just a minor addition, with the 7nm you can’t add anymore. You’re not getting the same level of performance improvements. It’s been capped. Because we are programmable across the board, 75% of all applications will be at the edge, whether it is autonomous cars or next generation of applications.
We’ve become an ideal platform where the cloud will move to the edge. Why? Because we have a low power footprint. We have the ability to process millions of packets at very high speeds. Because we draw very little power, you essentially are dealing with an underlying platform that can be used in the transition towards 5G or all edge-related workloads.