🦄 Making great presentations more accessible. This project aims to enhances multilingual accessibility and discoverability while maintaining the integrity of original content. Detailed transcriptions and keyframes preserve the nuances and technical insights that make each session compelling.

Overview

📖 AWS re:Invent 2025 - Building High-Performance, Secure networks for Hybrid and Multicloud (HMC209)

In this video, Brandon Wright from Equinix and Ranga Thittai from AWS discuss building high-performance, secure networks for hybrid and multicloud environments. Ranga covers customer use cases for AWS Direct Connect, emphasizing Gen AI workloads, migrations requiring hundreds of petabytes, and the benefits of dedicated connectivity with MACsec encryption. Brandon presents Equinix Fabric, a global software-defined network spanning 270+ data centers with 56 AWS Direct Connect on-ramps and five nines uptime SLA. He demonstrates Fabric Cloud Router’s performance, showing 9.5 Gbps throughput and 63ms latency between Seattle and Ashburn. Two case studies highlight a financial services company achieving four-hour deployment and an autonomous vehicle company transferring 500 terabytes daily. A seven-day trial is available on AWS Marketplace.

; This article is entirely auto-generated while preserving the original presentation content as much as possible. Please note that there may be typos or inaccuracies.

Main Part

AWS Direct Connect: Enabling Large-Scale Data Movement and Low-Latency Hybrid Cloud Connectivity

Thank you. Hey everybody, how’s it going? I’m Brandon Wright. I’m a Product Manager at Equinix, and I focus on our Fabric Network as a Service platform. I’m Ranga Thittai. I’m the Network Go-to-Market Head at AWS, and today we’re going to be talking about building high-performance, secure networks for hybrid and multicloud.

So super quick, just a quick agenda overview. Ranga is going to start us off and talk about some customer use cases, what they’re seeing around Direct Connect, as well as data movement considerations, Direct Connect, VPN, internet, and things like that, and what we see in terms of behaviors and looking at scale and time. And then we’ll look at the case for private hybrid multicloud, including some use cases and examples that we have on our side with case studies. And we’ll talk a little bit more about Equinix, as well as some of our products and the history behind Equinix and where we came from and where we are today. I’ll hand it over to Ranga.

Thank you, Brandon. Let’s start off by looking at the importance of large data movements as it relates to your go-to-market initiatives. We all know about Gen AI, whether it’s model training, foundational models. If you’re an ISV or an FM provider, you’re training your models. If you’re an industry enterprise, you’re probably looking at RAG initiatives. You’re building your RAG vector stores, and that requires a lot of data movements. You’re consolidating your data into centralized environments in AWS.

Talking of industries, it’s probably a cliché to state that industries run on data, but just to quote a couple of examples, automated vehicles, connected manufacturing, media digitization, media production in the cloud, all of them require large amounts of data to be moved to AWS, perhaps even on a continual basis. But across industries, perhaps the most common use case that we all come across is migrations. When you’re doing your complete migration to the cloud, you’re moving all your data estates to AWS.

It’s not surprising to see some of these requiring hundreds of petabytes of data to be moved to AWS within a matter of six to nine months. And to achieve that, you need reliable and high-bandwidth connectivity to achieve those objectives. Today we’ll talk about AWS solutions as well as partner constructs that enable these kinds of use cases, as well as network constructs that allow low-latency connectivity for distributed workloads as well as distributed data access. We’ll look at some of the characteristics of these solutions and also give you a couple of ways to try them out perhaps as well.

Let’s look at Gen AI. It’s a hot topic. So if you’re engaging in model training, Swami talked about frontier models this morning, which make you develop those models from scratch, specific to your industry domain. If you’re an ISV or a SaaS provider for AI, you’re also training your models or fine-tuning your models. If you’re just an industry player, you’re looking to leverage some of the foundational models but create your own RAG vector stores. That also requires a lot of data to be moved from hybrid ecosystems, hybrid environments, multicloud environments into AWS.

When it comes to inferencing, you could also be engaging in distributed inferencing. You could be looking at hybrid RAG if you’re compliance-oriented, or you could be considering hybrid EKS if you are looking to leverage distributed workloads across hybrid environments. In these cases, you’re looking at real-time interactions, and in these kinds of situations, getting a performant and deterministic network becomes important so that you’re delivering a consistent user experience through that network.

That brings us to Direct Connect. AWS Direct Connect is a dedicated network path into AWS. It avoids the internet, basically not subject to internet weather, and therefore delivers consistent performance all the time. It offers an availability SLA. So for critical projects like we spoke about, you’re guaranteed almost to have a continual available connectivity all the time to get your end-to-end project done, or if you’re distributing those workloads, mission-critical workloads, and you need that consistent performance, you’re getting that all the time.

There’s an extensive coverage. This offering is available in more than 30 countries in excess of 100 locations. Finally, it’s also encrypted in Layer 2 through MAC encryption. We call it MACsec. So if you’re in the compliance space, it’s particularly useful for specific use cases where despite that dedicated connectivity, you want your customers’ data to be moved in a compliant encrypted fashion as well.

Let’s put AWS Direct Connect in the context of those use cases that we discussed initially. So let’s start by looking at a simple 10 petabyte data move in one session. Later on, we’ll talk about hosted Direct Connect as a very simple turnkey construct that partners host, and they just carve out for customers like you. But a hosted Direct Connect offers a bandwidth of 25 gigabits, and with that you’d be able to achieve this transfer in one hour. But if you’re using a VPN connection over the internet to get the same thing, you’re probably about 5% progressed, assuming that availability stays 100%.

Now imagine using this benefit 1,000 times for a 10 petabyte migration project. That adds up that reliability and that ability to move that data fast for each of your little data islands as you prepare them and you curate them and you move them into AWS. That benefit adds up if you consider that. When it comes to distributed workloads and data access, AWS offers the shortest path. It’s also a dedicated connection. Therefore, it offers that performant connectivity that you require to give that consistent latency or response time, especially when it comes to workloads. We heard Swami talk about response times of these generative AI agents. You don’t want your network to be adding to that response time and latency as well.

All right, so how do you get your data and your distributed environments connected to AWS through Direct Connect? And that’s where our partners come in. Our partners offer that turnkey connectivity to get you connected quickly and easily based on your use cases. The connectivity models they offer are also elastic. They can also be customizable based on your vendor choices as well. With that, I pass this back to Equinix, our sponsors for the session, to take this forward.

Equinix Fabric and Cloud Router: Global Network Infrastructure for High-Performance Multicloud Interconnection

Thanks, Ranga. All right, team, let me talk a little bit about why Equinix. A little bit of the history of Equinix: we’ve been around for about 26 years. We started out as an internet exchange. Network service providers came through and said, hey, we’re trying to interconnect. Growth of the internet was exploding, and Equinix was that neutral interconnect provider. That business continued to grow, and soon after we had customers interested in connecting to public clouds. Public clouds as partners, and again, Equinix being a neutral interconnect partner, we’re able to provide connectivity to AWS and many other cloud providers, service providers, and network service providers. And then over time, we’ve grown to 11,000+ enterprises using our platform.

And then beyond that, it’s allowed us to create what we have today, which we call Equinix Fabric. This is a network as a service platform available across 270+ data centers in 36 countries, spanning 75 different metros and 6 continents. This is a completely global software-defined network. And of those, we have 220 or more cloud on-ramps, including AWS. 56 of those are AWS Direct Connect. You can see them, they’re a little off color there, squares on the map, and those are areas where we have direct on-ramps from our data centers in our Fabric network to AWS. And across our global network, we’re using more than 2,000 fiber and telco networks. All of that combined in providing that level of connectivity and reliability allows us to offer a five nines global uptime and reliability SLA.

As part of our Fabric network, one of the newer products that we’ve launched in January 2024 is our Fabric Cloud Router. This is a completely service-based solution. It doesn’t run on virtual appliances, there’s no licensing, there’s no other hardware required from the customer. It allows you to on-demand turn up instances in the metro that you want to deploy and create your connections from. It does this within minutes, and it’s fully distributed and supports natively high available connectivity, so primary and secondary data path connectivity that is completely independent end to end.

We offer various different packages and tiers depending on the performance and route scale requirements that customers have. But that allows us to offer multiple different, very economical options for Fabric Cloud Router. And then from a scalability perspective in terms of throughput, we offer everything from 50 megabit individual connections all the way up to 100 gigabit individual connections in an aggregate supporting greater than 1 terabit of aggregate throughput across that single cloud router instance.

As we look at different on-demand hybrid multi-cloud connectivity, we have customers that are deployed in multiple metros. They can deploy a cloud router, they can interconnect with their co-located infrastructure via fabric port. They can do that at up to 100 gigabit today, and then they can also connect to different cloud providers as well. One of the benefits here is that customers that may have traditionally interconnected between AWS and maybe a secondary cloud provider were doing routing in BGP back to a co-located router that they had. Maybe it was within an Equinix data center, maybe it was all the way back on-premises, which meant any connectivity that was routed between the two cloud providers had to effectively hairpin back to wherever that router was. And that’s where Fabric Cloud Router comes in and provides extremely low latency interconnect as we have local on-ramps with multiple cloud providers across our metros.

Looking a little bit more into the performance, on the right you’ll see some data that I’ve pulled. This is some testing that I’ve done a few months ago, but a couple of key points to call out here. On top of the fact that the Fabric Cloud Router is completely private and operates across our Equinix private network, we do also have what we call IP WAN or it’s an IP VPN network. It’s a true multipoint to multipoint network, and I can interconnect my Fabric Cloud Routers across different metros within, let’s say, North America. I can also extend into EMEA and APAC as well. So I have both regional and global network constructs. I can set the speed or throughput that I want on those, and in this particular example, I’ve got a cloud router that’s deployed in Seattle and I’ve got one deployed in Ashburn.

I’m connecting to AWS on the West Coast, and I’m connecting in this case to GCP on the East Coast. I have 10 gigabit virtual connections. So this is an AWS Hosted Direct Connect, as well as Google’s counterpart there with regards to cloud interconnectivity. And then I’ve got two virtual machines, one on each end, and going from west coast all the way to east coast, between 63 milliseconds on average. A couple of things to call out here. When we’re doing interconnectivity between these metros and going across our private network, rather than going across the internet, we effectively have four predetermined hops. I have my Direct Connect gateway, I’ve got my cloud router, my second cloud router on the other side of the country, and then my Google Cloud router on the other end of the connection.

Beyond that, obviously there’s the endpoints or the IP addresses of the actual VMs, but it’s extremely predictable. Whereas if I’m going over the internet, I can run two different trace routes and I may take completely different paths. It may be somewhat unreliable or essentially best effort, right? So there’s a lot of predictability in terms of performance and jitter. So for latency sensitive applications, even across longer distances, we see a lot of success there as well with Fabric Cloud Router in our Fabric network.

Taking it a little bit further down, customers should expect to see line rate across this connectivity. There’s still layer 2 connectivity there, and with the Fabric Cloud Router, we’re just building layer 3 connectivity on top of that. So in this particular example, you’ll see, if you’re familiar with iPerf, these are two very, very small VMs with my only criteria being they could do 10 gigabit out. So the VM size, number of vCPUs, frequency and amount of memory is extremely limited. So you’ll see I ran 100 parallel threads with TCP and changed the congestion algorithm because I was going across either end of the country, and it was completely saturating my virtual machines. But we were able to achieve 9.5 gigabit per second on one end and 9.26 on the other end as well.

And then on the bottom there when I’m using UDP, it’s a lot less, no congestion there. And you’ll see that I effectively hit very, very, very close to what I would expect to get line rate with 63 milliseconds of latency. And you’ll see on the very, very bottom, actually, I’m not running multiple parallel threads there. And you’ll see 99.89 gigabits sent, and I’m only getting 4.92 back, and I’m simply running into single flow limitations with connectivity on the return path. So that’s not a limitation on the cloud router. It’s just the AWS networking single flow limitation. If I was to run parallel threads there, we could saturate the whole thing, no problem.

Real-World Success Stories: From Four-Hour Deployment to 500 Terabytes Daily Data Transfer

The last thing we’ll kind of talk about is some customer case studies. The first one here is going to focus on speed and agility, and then the second one we’re talking about is raw scale, right? So we had a customer approach us sometime back. This is a completely new customer. They had heard of Equinix. They weren’t currently using Equinix. They came to us and said, hey, we’re deployed in AWS on the East Coast. We recently acquired a company. We work in the financial services industry, particularly in automotive loans, and one of the products that we participate in, the dealers will query out when an individual is looking to purchase a vehicle, and they’ll reach out to multiple institutions,

and it’s almost kind of a race to see who responds back first. It goes down in a list, and what they were aiming to do is move higher up in that list in terms of how quickly they can respond. If that person was approved for the loan, they can respond quickly, get closer to the top of the list, and therefore may have more success with the dealership selecting their institution.

This particular morning, I jumped on a call with a salesperson. They said, hey, I’m talking to somebody new. Can you help me out? I said, yeah, sure. So I talked to somebody for about an hour. This was about 8:30 in the morning. Then by 12:30 in the afternoon, he had reached out and said, yeah, I got it deployed, it works. So within four hours, somebody who had never even had an account jumped in, went through our process, and in this case, they went through our marketplace, which I’ll share about in just a moment as well. Effectively, met us for the first time, I gave him some direction, and then he went off and running. Later that afternoon, he was extremely happy with his ability to interconnect AWS with Azure.

The second case study here is an autonomous vehicle company with some huge requirements in terms of data that they’re transferring. Autonomous vehicles collecting data periodically, multiple fleets of vehicles across the country, effectively totaled 500 terabytes a day that they needed to transfer into a mixture of S3 and other public clouds. One way that we helped solve this for this customer across the country was having a physical presence in the data centers in the regions where they had their autonomous vehicles and then utilizing AWS Direct Connect with multiple 25 gigabit circuits, virtual connections, to move that data into AWS. We were able to successfully do this for the customer. Even with the ability for that fleet to require moving, because of the number of data centers that we have across the country, it’s worked extremely well for that customer.

Finally, as I mentioned a moment ago, we do have a seven-day trial of our Fabric Cloud Router that is available on AWS Marketplace. If you’re interested in multicloud connectivity, hybrid multicloud, and want to give it a try, it’s our full standard paid version, but it’s free for seven days, and it comes with two 200 megabit virtual connections that you can turn up. If you’d like to use a smaller speed or a lower speed as well, down to 50 megabit, you can do that as well.

Last thing I’ll mention is I’m happy to answer questions. We have a booth that’s actually right behind the pavilion here, booth 335. If you’d like to come by and ask any questions, otherwise, I think we’re good. Thank you and please do take a moment to give us survey feedback as well. It should be in your apps. Thank you very much. Thank you.

---

; This article is entirely auto-generated using Amazon Bedrock.