Intel Corporation (INTC) Earnings Call Transcript & Summary

Hello, and welcome, everyone, to the Intel Network Builders webinar program. Thank you for taking the time to join us today with Lanner and EnterpriseWeb for a presentation titled Building a Dynamic Network Infrastructure with Open, Programmable and Scalable Building Blocks. Before we get started, I want to point out some of the features of the BrightTALK tool that may improve your experience. There's a Questions tab below your viewer, and I encourage our live audience to please ask questions at any time. Our presenters will hold answering them until the end of the presentation. Below your viewing screen, you will also find an Attachments tab with additional documentation and reference materials, including a number of websites and documents mentioned in this presentation. Finally, at the end of the presentation, please take the time to provide feedback using the Ratings tab. We value your thoughts, and we'll use the information to improve our future webinars. Check the Intel Network Builders website partner webinar channel to see what is upcoming and access our growing library of recorded content. In addition to the resources you see here from our partners, we also offer comprehensive NFV and SDN training programs for the Intel Network Builders University. You can also find a link to this program in the attachments tab as well as a link to the Intel Network Builders newsletter. Intel Network Builders partners have been working to accelerate network innovation by optimizing their solutions on Intel technologies. These industry leaders are recognized in our Winners' Circle program. Lanner is a titanium partner, and NoviFlow is an INB Winners' Circle member. Learn more about our INB Winners' Circle program by clicking on the link on the Attachments tab. Today, we are pleased to welcome Babu Peddu, Senior Product Marketing Manager at Intel; Sven Freudenfeld, Chief Technical Officer at Telecom ABU at Lanner; and Kevin Austin, Director of Product Management at NoviFlow. Welcome, Babu, Sven and Kevin, and thanks again for joining us today. Now I'll go ahead and hand it over to our presenters to get us started. Babu, all yours.

Thank you, [ Sam ]. Hello, everyone. Good morning. Good afternoon, and good evening, depending on where you are located. Thank you all for joining this webinar. And the topic for today is building a dynamic network infrastructure with open, programmable and scalable building blocks. So let's go ahead and get started. The agenda is I will discuss the industry trends and the Intel vision, and we'd also talk about the data explosion, networking challenges, and then I will introduce you to Intel Intelligent Fabric vision. We'll also discuss about the Intel Tofino form factors, and then I will dive into the interaction of Segment Routing v6 version. Upon that, my partners from Lanner and NoviFlow, they will deep dive into SRv6. So industry trends and vision. In today's world, service providers, cloud service providers and enterprises, they are considering programmability in their networks. The reason for that is programmability offers a bit of customization. In addition to the customization, it also offers low-cost resilience and optimizations for all the telcos and enterprises. And then with the advent of new technologies like 5G, AI/ML, IoT, edge computing is transforming the way how the data is being delivered to billions of devices across the globe. So at Intel Connectivity Group, we have purposefully created to deliver fully programmable networks and end-to-end connectivity. At Intel, we have the right vision, right technology to partner with all of our ecosystem partners to help and build the next-generation programmable networks. This is an excerpt from Gartner. Gartner projects that by 2025 there will be more than 50% of data that will be manual and processed at the edge. With that keeping in mind, it is imperative to have more compute power, storage and analytics closer to the edge, not at the data center but closer to the edge. So is the current infrastructure geared towards that? I don't think so. And then we have seismic shifts. Web services companies like Facebook, Amazon, Alibaba and LinkedIn, they are serving billions of users; and then on the other hand, there is a lot of demand for high-quality video. So -- and then there is also demand for AR, VR metaverses. And then this is all resulting in demand for multiterabit performance. And then changing and evolving workloads, so there is -- due to the growth of cloud and need for workload scaling, there is a lot of transition from on-prem cloud to hybrid cloud, and then they are instantly making decisions to move to a private data center. With these kind of changes and evolving workloads, there is demand for programmable networking as well. So let's look at some 5G networking challenges. Globally, there is significant growth in mobile data traffic. With 5G, it's going to be even more. This can be attributed to 3 main drivers: one, improved device capabilities; two, an increase in data-intensive content; and third -- three, more data consumption due to continued improvements in the performance of deployed networks. Many service providers are now on a shift to cloud data architectures with container-based processing microservices, network orchestration and automation. These changing workloads form the need for artificial intelligence, network optimizations as the industry migrate to cloud native and distributed scale-out architectures. These distributed architectures, they designed to put more emphasis into the hands of end users. And end-to-end security and visibility approach is also a consideration but not a backup at all. So we cannot solve newer problems with older techniques. To also say that, we cannot squeeze more throughput. All these technologies require more and more throughput. We cannot squeeze more throughput from the existing silicon. Bottom line [ for us ] for 5G services, networks needs to be more smarter with increased bandwidth. We will need more network intelligence moving forward. So how can we solve all these problems that I discussed until now? I wanted to introduce you to Intel Intelligent Fabric architecture. This architecture brings together a set of Intel portfolio that includes silicon photonics, Tofino, intelligent public processor, IPUs, Intel Ethernet, Intel Xeon CPUs, Agilex, ESX, Habana, Intel Xe. So all of this portfolio in fact, supports the open standards that brings together P4 programming language, IPDK, SONiC, that is the networking operating system, eBPF and data plane development kits. This portfolio and open centers together provide better resiliency, optimization, massive throughput for all the future networks. Here, I wanted to share the 2 form factors that are on track today and we are selling. Tofino, it comes with a 16-nanometer kit, and it provides up to 6.4 terabits per second with 25-gig SerDes. And then the second form factor is Tofino 2 with 7-nanometer. It provides up to 12.8 terabits per second with modular chip design and 56-gig SerDes. So all of -- these 2 form factors together, they provide intelligence with P4 programmability, AI/ML acceleration, highly secure. And then performance, as I said, it goes up to 12.8 terabits per second throughput on a 56-gig SerDes. There is a lot of core optimization use cases that can be leveraged here as well. On the visibility and control, we offer deep monitoring, deep visibility into the networks to remediate any problems and then provide edge-to-cloud real-time telemetry, that we can also support enhanced condition as well in these things. So the programmable switch ASICs enable composability, efficient pipelines by restricting the unused or unnecessary protocols as well. They also provide the ability to implement new or customized protocols as service power requirements keep changing ever and ever again. All of this will result in an increased performance and power optimization for hyperscale and high-performance computing use cases. Now let me also introduce you to Intel Tofino Expandable Architecture. As I spoke earlier, for high-performance computing and other web scale use cases, which require massive performance, they can leverage Intel Tofino Expandable Architecture. This architecture combines Intel Tofino switch ASICs and Intel CPUs with one or more accelerators like Intel FPGAs and Intel IPUs. On the use cases we support Layer 4 server load balancing acceleration. On the security side, we support firewalls. We support DDoS mitigation. And then on the telco -- for the telcos, we support broadband network gateways, AGFs and UPFs. For the [ parts ] of service providers is more cloud gateways. And then we support network functional virtualization too, with the compute storage and analytics. These are the main use cases. And let's look at these benefits actually for the Intel Tofino Expandable Architecture. Massive performance and scale, flexibility, open center support like P4 and everything, all of these 3 benefits will result into a larger benefit that has lower CapEx plus OpEx that will again result in lower TCO. So this is what the Tofino Intel Expandable Architecture will provide. Now let's talk a little bit about the segment routing. The Internet protocol version, which is the IPv6 adoption, is growing at pace. Else -- somewhere, Google have also indicated that a significant number of users are reaching Google over the IP6 version, that is the new IPv6 protocol. The Internet of Things, the Internet of everything relies on IPv6 as it will provide connection for millions of IoT devices in the future. IPv6 is also a key and significant enabler to [ offer a ] segment routing concept as it provides reachability to SRv6-capable nodes. So what is segment routing? Segment routing is a new way of doing source routing where the source selects a path over a network, placing an order list of 128-bit IPv6 addresses into the header of an IPv6 packet. It is unlike IPv4. So the SRv6 will be the next-generation torchbearer protocol that combines both segment routing and IPv6. Utilizing the existing IPv6 forwarding technology, SRv6 implements network programming through flexible IPv6 extension headers as well. SRv6 has been deployed in almost, I would say, 8 public large-scale commercial networks including SoftBank and many other version -- other implementations. Also, there are many hardware implementations that are supporting SRv6. These implementations span across custom silicon, merchant silicon that will include Intel and many other factors. This is also backed by open source that includes P4 and SONiC, which is very networking oriented. So why SRv6 actually? So Segment Routing v6 is a software-defined networking solution, which provides powerful programming ability, powerful customization, and it also provides great flexibility for steering the traffic. And is it difficult? No, not at all. It is extremely easy to deploy. It uses IPv6 forwarding plane and requires -- and doesn't require the complex multiprotocol label switching protocol scheming. It doesn't need a lot of hardware support for forwarding as well. From the Intel side, we have the SRv6 components that include Intel Xeon, Intel Tofino, Intel Ethernet and the Intel P4 programming language. So folks, with that, I will hand over the baton to my colleague, Kevin. Kevin, please go ahead and take the ball.

Sure. So the question is what is SRv6. Babu covered that a little bit. Before that, I want to take a step back and just bring up a couple of things. Babu had mentioned seismic shifts. This truly is seismic shifts going on in the telco right now and in their networks, as they're dealing with the move from physical to virtualization or containeration -- containerization, decentralizing the network and pushing everything out to the edge and the increase in traffic. These are huge, huge shifts that are going on, and really what the carriers need are programmable networks. And SRv6 is that new programmable network technology that's going to carry them forward. It's already being adopted in a lot of the major carriers. And the IFP processor from Tofino allowed us to leverage SRv6 in its initial generation. If -- a lot of people probably remember back in the VXLAN days, it took a couple of silicon iterations in order to get full VXLAN support. With the IFP processor, we're able to build support for SRv6 into the existing silicon. So it's a great canvas for us to do that. But at its basic, SRv6 has 3 different components to it. It's doing encapsulation. You have a head end that's going to encapsulate the path of the packet and the services of the packet into the packet header called the SRH. The second type of device you have is in a segment or transport. So it's just a transit node that goes from point A to point B or is just one of those nodes along the hops. And finally, you have the endpoint that's going to remove the SRH header and move the packet to a non-SRv6 network. So couple of great things about SRv6. It's used for traffic engineering. So we talk about network slicing. SRv6 is one of those mechanisms that you can implement in order to get to network slicing. Allows you to implement Layer 2, Layer 3 VPN services and also do service chaining, which we'll talk about. So why do -- why SRv6? It reduces the cost of delivering network services. So that -- number one, it doesn't require special hardware in order to use SRv6. So it's a lower cost than some of the other solutions used in the past out there. It allows you to do service programming. So you can start to decouple the transport network from the services run within that network. It's a better scaling of the network. So as you need more capacity, more bandwidth, you scale the number of nodes and the services that you implement. It simplifies the network. It reduces the number of protocols that you need, and it's easy to deploy. So you can deploy SRv6 over an existing IPv6 network and not all of the -- not all of the network devices have to be SRv6 aware. And that's a big difference between SR over MPLS and SRv6. But also, it's tenable to add new network functionality, like the SR proxy that I'm going to talk about, a service chaining and 5G UPF. So our product, NoviFabric -- and my partner, Sven from Lanner, is going to talk about this a little bit more. But we're addressing 2 problems with this joint solution. And our product, NoviFabric, sits on top of the Lanner platform, and we virtualize the services that a carrier would run typically on dedicated physical appliances. We now virtualize that on a set of servers within a chassis. And using the IFP processor, we create SRv6 service proxy. So now we bolted these sets of services and created a services domain, and we act as the orchestrator and enforcement point. So this takes away the lock that you have from the transport network and what the needs of the guys running the services for the infrastructure. Now it's completely 2 separate domains. And we support a mix of different types of services because not all of these services are SRv6 aware, so they need to be proxy. They need to have some intelligence to deliver traffic to them. And each of the services, it can be hardware, NFV or a mix and match of them, but this is the way that you build the service on to an SRv6 network. And as you see here, we've created this domain. And one of the things within SRv6 that we've created is this END.SC. So it's this new function where we can take -- it acts as a binding SID and isolates that service chain from the rest of the network. So when we receive an SRH header multiple different service chains can be programmed into SRH header that we execute. So if there's a certain number of services, certain chain that you want to evoke, we can evoke multiple chains through that SRH header. And then the second thing that we do here is you see that there's multiple appliances attached. We solve the other problem of scaling those applications because we included a load balancer as well with the proxy. So you're able to expand firewall services by firing up more virtual machines or adding another server blade. So it allows you to scale and insert services seamlessly into the network. Now there's 3 types of applications that we've identified, and we call them Type 1, Type 2, Type 3. And the Type 1 is this is where the application, be it a firewall or DDoS, it's aware of IPv6 or, more accurately, SRv6, and it can parse that SRv6 header, and it understands it. And it's another addressable device in the network, and it will parse that header. It will shift it to the next and send it off to the next hop. And that's an SR-aware application. And there's not that many SR-aware applications out there today. That too is an SR pass-through. And this is where the application really doesn't parse the header. It doesn't need to understand the context what the packet is. And so it just passes it through unchanged. And the third type, and this is the most popular type of application out there is, SR unaware. It doesn't know how to parse that packet. If it receives a packet with an SRH header, it would -- it wouldn't know how to classify that. So one of the things that we've done in that proxy is we will remove that header before sending it to a Type 3 application. And when we get the packet back from that application, we'll reinstitute the header onto the packet and then send it off either to the next application in the chain. We'll send it off to the next hop. And with that, I think I'll turn it over to Sven to talk a little bit about the hardware platform.

Sure. Thank you, Kevin. But before we go over, I just want to reference the poll question we have in our webinar, in particular, regarding the SRv6 use case. There is a major trend in the market, and service providers are considering deploying SRv6 in the network. We've been working with at least 2 major service providers in a live deployment, optimizing or leveraging SRv6. And the poll question will help us also to get a bit more insights on the status of deploying networks based on SRv6. Having said that, the foundation to be able to provide network programmability is the underlying infrastructure. And the underlying infrastructure is really a combination of network programmability, so providing and leveraging ASICs in the market, which are -- enable programmability is one of the key aspects. And of course, because we are talking about deployment in some of the edge location, networking and computing going hand in hand, so where we have connectivity and compute resources right at the edge location. And typically, networking, it's much more dynamic at the edge location than it is at the core network because traffic types coming in into the edge location is very diversified. It could be IoT traffic, could be voice traffic. It could be any other critical or encrypted traffic, wherefore, the network programmability is crucial at the edge location. And therefore, whenever -- programmability using the Tofino as a switch fabric in front of the compute nodes is a classic use case. It allows us to scale segment routing into segmented, the traffic or also sliced traffic in the network right at the edge. And of course, in order to have security functions and other VNFs, virtual firewall, virtual DDoS and other vital network functions right at the edge, we need the compute resources. So therefore, we create a carrier-grade appliance, which is consuming very low footprint, consuming very low power and is your typical telco characteristics, where we can scale the SRv6 framework and become a proxy towards the transport layer of SRv6. So that's part of the scalability. We're leveraging the most common elements in the platform, including backplane connectivity using the Intel E810 networking devices we're providing. We are using the latest Intel 4 generation scalable processor, also known as Sapphire Rapids. So it's providing a lot of flexibility and with the new technology on the Intel -- fourth-generation Intel, we actually have even much better optimization for managing the workload for a compute node. And last but not least is visibility. So these devices are located in multiple locations, remotely at 6,000 or 7,000 locations all over the country. And they are very hard to accessible -- to access. So therefore, providing visibility and better visibility is a key aspect for that. And just the switch asset of Tofino architecture allows you to leverage in-band network telemetry. But it's not only on Tofino. It's also on acceleration modules, which can provide and leverage P4. So fundamentally, on the platform level, what do all the elements look like? It's the telco-grade solution, where we built the platform, integrating networking, computing acceleration and timing into the platform. So these are modular platforms, which means that once deployed in the network, it will not lose its value. It can update towards the next-generation Intel Xeon platforms by updating the compute nodes. We have the fully programmable switch ASICs built into it, which is also containing timing, network timing or IEEE 1588 for some of the use cases. And then, of course, we have a centralized management to manage all the different elements. Just imagine you have a whole server farm built into this platform and to manage all the resources, you need some sort of a framework to do remote management, remote diagnostics and also to optimize the workloads in terms of networking and compute workloads. So that's all combined. It's really the foundation to the deployment of SRv6 and edge location becoming a proxy towards the SRv6 framework. We actually evolved this platform also. We have a much more compact solution built into it, which is also allowing you to maximize the accelerator framework where we have standard form factors on PCI Express form factors or even the open compute modules into the platform. So that means it's even much more flexibility anywhere, where we can use it as use cases are allowing into a mobile infrastructure like the centralized unit or even the distributed unit in the mobile space. But the networking aspect remains the key for edge location. And since we integrated the network switches, providing SRv6 into a network infrastructure, we have much more scalability and flexibility. So with that, I'm handing it back to Kevin to give a bit more granularity on the software elements to manage and to scale the platform. Thank you, Kevin.

Sure. Thanks, Sven. So what we see here is kind of the different layers of this platform. This is with the NoviFlow SRv6 service engine components with our NoviFabric on top of the Lanner platform. And essentially, the customers that use this platform, they have 2 goals. One is I want to virtualize my infrastructure. And that requires a lot of getting down to the real high-performance nature of the platform. And this box is unique in that it's designed for the data plane. So each of those server blades, and there are 6 over blades in this chassis, have 4 100-gig links, 2 to each of the switches from each NUMA node. So each NUMA node gets an independent link to switch A and to switch B. So that's the first thing. The second thing is the scalability. So we have the fact that you can insert this as a service anywhere. We have the fact that you can turn up additional services in the box that you need, be it firewall or some other application. And we can load balance across to it because we've got a built-in load balancer. But we also have filtering, packet sphering, servicing. All of that is built into this product. Now when -- the second point that I wanted to make is that a customer typically has their own cloud stack or their own edge stack that they would like to use. They've got existing relationships with OS vendors. And we will work with those vendors and create the right edge stack for our customer and do the integration work there, so very flexible, lots of tools that you get to use here with this product. We also have NoviAnalytics and NoviDashboard, which completely monitors the system from the hardware layer up through the networking and compute layers. It's -- and visualizes everything for you for operational health, so real comprehensive edge platform here between Lanner and NoviFlow. And as I mentioned, this is a great way to seamlessly insert services via SRv6. So the Lanner platform is basically a CO in a box. And using our software, the NoviFabric software, you can insert these services anywhere within the network from all the way out to enterprises to global data centers. And there's 2 major use cases, and we've talked about this quite a bit. But the scaling of the network security services, so as you move from the centralized appliances, purpose-built appliances out to the edge, the amount of bandwidth that you have to handle is, of course, less but you still have to put more of these devices out at the edge. And sometimes you'll need to scale those services across maybe a server, a couple of NUMA nodes, a few servers, and to get the performance, you need a load balancer. Now it's not really cost effective to put a load balancer out at every edge site. But we've built that in to the Tofino IFP processor, so you get that with the platform. You get that along with SRv6. You get that extra HA that you get from the load balancing across multiple resources, so has a lot of functionality out there that traditionally would take maybe a packet broker or switches and a load balancer, now collapsed into one device. And we typically see this being deployed on the mobile backhaul or Gi-LAN. So that's where you're going to want to be putting your services depending on which side of the network that you're protecting. And this allows the customers by using this platform to save a lot of money in that by integrating all the networking that you need, all the compute, all the storage, all of the load balancing and the service proxying, it really gives you a cost efficiency comparison. And Sven, I'll turn it back over to you to maybe talk about this slide a little bit.

Yes. So to sum it all up, why is that a valuable solution for SRv6 and what makes it so special? It's basically where we -- by leveraging with Tofino ASIC, we actually eliminate the need for dedicated appliances, in particular, to do load balancing, as Kevin said earlier, and other functions in the network. So we have it fully integrated. So we -- the programmability and the P4 capability of this switch ASIC allows us to move some of the applications, which are typically commercially proprietary or in that sense also dedicated for one workload. We actually can scale that into our platform. And the bottom line here is that we're getting the same or greater functionality with much less efforts, meaning that the cost is reduced because we can reduce the footprint, less real estate, less power in -- and less maintenance and operations cost because we have all the tools that built into it to manage all the elements. And then rereducing the space also by -- there's efforts for cooling. As I said, less power consumption, less rack space. Some of these locations, the edge locations, they don't have a luxury of operating in a hot aisle and cold aisle environment where we -- what you -- what we know from a data center environment. But because of that integrated combination of computer networking, we have much more flexibility. And that means that everything can be preconfigured, sent on site. You reduce the [ track roll ] to configure, maintain and operate it. And nevertheless, just thinking about all the networking ports, transceivers, cable, fibers and so forth, because we're using that integrated approach, it's eliminating the need and one less point of failure by configuring these edge locations. And again, that is a big cost advantage rather than having with traditional purpose-built or dedicated server, top-of-rack switches and load balancer approach. So it makes a big difference by managing the network and much more efficient. So -- and it's all thanks to the technology built into it. So from a programmable switch ASIC using the P4 framework, deploying SRv6, which is standardized, it's been deployed in service provider networks. It becomes more obvious that it's a better way of slicing the networks using Segment Routing v6. So with that, I'm handing it over to Babu.

Yes. Thank you, Sven, and Kevin, very insightful and a nice presentation. So folks, I wanted to close out with a P4 emphasis. So what is P4? It stands for programming protocol-independent packet processors. It's an open source and domain-specific programming language, where we have built out leveraging the Tofino. So why use P4? It provides greater visibility. P4 allows users to program and rules to the forwarding devices. It can, for example, create a tag for each of the packet as it passes through a router or a switch or any networking device, I would say. By doing so, it lets network engineers get a potentially unprecedented level of visibility into the routing parts of packets to determine network latency, create more optimization, create more customization, et cetera. So with P4 and network intelligence, customer customers can customize data flows so they can rapidly innovate and differentiate the next-generation workloads. So Intel has a P4 studio environment that can be leveraged, which many of our telco, Tier 1 service provider customers are leveraging already. We got great input on that so far. So that's all. And with that, [ Sam ], let me pass on the baton to you. Thank you all.

Yes. Thank you so much, Babu, Sven and Kevin, for that great presentation. We do have several questions that have come in from the audience while you were presenting, so let's get those started. The first question asked, are there any labs that we can access to experience the P4 programmable networks?