Introduction

The essence of today’s technological progress is the explosive growth in data transfer, with global internet traffic projected to double within the next 3–5 years. Almost daily, we hear about new data centers being deployed — from industry giants to smaller players — all investing heavily to expand capacity. But simply building more infrastructure is not always the best solution. To truly keep pace, we need smarter strategies that handle data more efficiently. One of the most impactful approaches is to bring data closer to the end user, a shift that not only improves user experience by reducing latency and buffering but also eases the cost and strain on ISP core networks. After all, in a city of 10 million people, it’s unrealistic to imagine every Netflix stream traversing the ISP backbone — at national scale, it becomes unmanageable. This is where Edge Data Centers come in: facilities designed to sit closer to users, enabling faster content delivery while reducing the load on centralized infrastructure.

In this blog, I’ll explore a case study of a CDN-based Edge data center and highlight the features and capabilities a data center switch must have to meet the unique demands of an Edge Data Center.

While there could be many other important use cases for the Edge DC like MEC (Mobile Edge Computing), IOT and Smart Cities, Gaming, Retain experience etc but I will pick them up in other case study. In this blog I will focus on CDNs. Over the last decade, video streaming has transformed from niche to mainstream, driving enormous growth in Content Delivery Networks (CDNs). Services like Netflix, Amazon Prime Video, and YouTube rely heavily on CDNs to deliver high-quality, low-latency video to billions of devices worldwide. Let’s break down how CDNs make money, how ISPs fit into the picture and how EDGE DCs are becoming more crucial than ever for on demand streaming applications.

By the way, there is something called Far Edge Data Center as well. For our discussion purpose we will consider far edge as even smaller footprint than Edge, and thats the actual definition as well in my opinion 🙂

Lets see what are CDN and How do they Make Money

A CDN (Content delivery network) is a global network of servers designed to deliver content quickly by caching it close to end users. The business model revolves around selling performance, scalability, and security to content providers.

Revenue Models

• Bandwidth / Data Transfer Pricing: Websites pay per GB delivered. Example: $0.05 per GB in North America, higher in Asia. Large streaming platforms pay millions monthly.
• Subscription / Tiered Plans: Flat monthly fees for a quota of traffic. Example: $200/month for 5 TB.
• Feature-based Pricing: Add-ons like DDoS protection, Web Application Firewall (WAF), video encoding.
• Enterprise Contracts: Long-term deals with OTTs, gaming, and SaaS providers.
• Freemium Upsell: Free CDN for small sites, but charge for premium features.

Example: Netflix in a 10M Population City

(Even though for 10M city there would be a good size DC deployed by ISPs/public cloud vendor etc .This example is more for understanding the CDN costing)

• City population: 10 million
• Netflix penetration: 20% or 2 million
• Average viewing: 30 hours/month per subscriber
• Average streaming bitrate (mixed HD/4K): 5 Mbps → ~2.25 GB/hour → ~67 GB/user/month
• Total monthly traffic = 2,000,000 × 67 GB = 134,000,000 GB (≈ 134 PB/month)
• Typical commercial CDN prices (per GB): $0.01, $0.02, $0.03, $0.05 (range from aggressive enterprise to retail) Lets consider 0.03$ retain price for this example.

Raw CDN Egress Cost (no ISP caching)

Now lets calculate Netflix cost to CDN provider for a 10M city. We are calculating just “egress cost” which is pure bandwidth without any services, CDNs put many add on features and security services that increases the cost.

Monthly egress cost =

• CDN rate ($0.03/GB) x per user yearly data read(67GBx12)x total user(2million)= ~$48.2M / year

And how much Netflix may need to pay for a subscriber to CDN (monthly) = monthly cost ÷ 2,000,000 subs:

• monthly CDN cost (48million/12months)= 4 million
• 4 million/total subscriber(2 million)= “2 dollars/subscriber paid by Netflix to CND“ Impossible!!!!!

Now if we consider netflix subscription cost in India, 2 million subscriber will pay almost 12 million/month -> 144million/year to Netflix and seeing that CDN cost itself is close to 50 million/year , Its 1/3rd of their revenue from the city,which I am sure is not a good business because there will be so many other cost that will come apart from CDN. So this will leave netflix operating in loss, which is not true.

So , math is not mathing well 🙂 Lets see what are the options that netflix uses to keep the cost of CDN low.

What are these option?

Public cloud CDN like CloudFront can be a good option and its a best option when your streaming application is sitting in some of these public cloud. ISP caching is a very effective option as normally ISP will not charge or charge very less to host the CDN for the big providers like Netflix because they save on the core bandwidth, think that ISP will not have to route that much of media through their core network which is served by the ISP caching Netflix server. Also the content provider support this initiative by providing appliances and infrastructure for free.

More info here : https://openconnect.netflix.com/en/appliances/?utm_referrer=https://www.google.com/

ISP Caching and Monetization

ISP caching means putting content servers (caches) inside an ISP’s network. Instead of fetching Netflix or Prime content from faraway CDNs, users get it from a local cache.The main benefit of this approach is that ISP is everywhere and they already have some or the other infra sitting, real estate acquired, people streamlined, government regulations sorted, customer base built and what not!!!

How ISPs are solving the bigger problem with the simple solution with Edge Data centers.

How It Works

• Content providers (Netflix, Google, Amazon) ship caching appliances (e.g., Netflix Open Connect Appliances).
• ISPs host them in their data centers.
• Popular content is preloaded, reducing duplicate upstream traffic.

How ISPs Monetize

• Transit Cost Savings: Less backbone traffic = millions saved.
• Better User Experience: Lower buffering means happier customers.
• Bundling: Easier to partner with Netflix/Prime for subscription bundles.
• Traffic Engineering: Frees up backbone capacity.

Netflix Open Connect Architecture

Netflix built Open Connect, its own private CDN, rather than relying solely on third-party providers. They designed custom Open Connect Appliances (OCAs) — servers preloaded with Netflix content — and shipped them to hosting partners including ISPs, cloud regions, and internet exchange points (IXPs). The brilliance lies not just in the hardware, but in the strategic placement of these OCAs. Thanks to this model, it’s estimated that over 90% of Netflix traffic never leaves your ISP’s network. The level of planning, coordination, and flawless execution behind this is nothing short of remarkable. Hats off to Netflix and this whole ecosystem!!

Components

• Open Connect Appliances (OCAs): Physical cache servers placed in ISP networks or IXPs.
• Push Model: Popular shows preloaded onto OCAs (e.g., new season episodes delivered overnight).
• Hierarchy: Regional OCAs at IXPs → Embedded OCAs in ISPs → End users.
• Control Plane: Netflix app selects the best OCA using DNS + BGP routing.

So far , I have tried to define and break down the problem and solution for CDN provider’s perspective. Now lets see where the DC vendors comes into picture and how they are keeping themselves ready for Edge and Far Edge DC requirements.

What is Cisco Data Center switching role in this.

The Role of Data Center Switches in Edge Data Centers

We’ve established that Edge Data Centers are one of the most effective solutions to the growing challenge of internet traffic. But to make them work, the foundation lies in the data center switch.

Take the example of Netflix’s Open Connect Appliances (OCA): servers preloaded with popular content and refreshed with new titles overnight or on demand. These servers are housed in edge facilities, positioned close to users. Their primary job is simple yet demanding — deliver HTTPs-based video streams at scale, with minimal delay, to millions of viewers.

Here’s where the data center switch comes in. Every stream request flows through it. The switch interconnects the OCA servers and hands off the traffic to ISPs, ensuring that terabits of data per second can move smoothly without becoming a bottleneck.

The key requirements of such a switch in an Edge Data Center environment are:

Lets try to understand each of these points.

• High Throughput & Port Density – You need a switch that can support high number of ports and provide throughput in multi terabit. I will take an example of Cisco GX2 Data Center switch 9364D-GX2A. The device has 64×400 in 2 RU and provides 51.2T of bandwidth.

• Ultra-Low Latency – Ultra low latency generally is considered anything below 5usec latency but lower is better. We can understand it more practically by an example of a video stream. Netflix says that as long as the streaming server and the user is within 50- 80 ms of round trip time the video will play smoothly. And generally when we consider round trip time, the time take by switch to process the packet is almost negligible. Considering that these POP(point of presence) are near to user hence I would consider less number of hops as well. So I would say if the switch is having latency anywhere below 5 micro seconds it should be fine and good that we have 9364D-GX2A providing ~1micro seconds of latency. Lets calculate how much traffic this POP is expected to receive for 10 million city.

City population: 10 million
Netflix penetration (20%): 2 million subscribers
Peak activity (30%): 600 K concurrent streams
Average bitrate: 5 Mbps per stream (mix of HD/4K)
Peak traffic = 600,000 × 5 Mbps = 3 Tbps

So really looking at this calculation, it looks like the DC switches we have are far more capable to handle it.

The switch should also have sufficient buffer to handle any micro bursts as well, like GX2A will have 120 MB on-die buffer to handle it.

• Efficient Flow Handling – Millions of simultaneous unicast HTTP sessions must be managed without packet loss. This one is very important that the network and server cluster should be able to handle the tcp flows correctly. Also inside the POP there can be a requirement for a L4-L7 device insertion and HTTP session handling when there is a L4-L7 devices in between becomes a bit challenging. That is where intelligent service chaining can come into picture. Cisco ACI with Leaf switches like 9364D-GX2A is capable of linline service chaining. And on top of that you need to have intelligence like while service chaining for a particular flow the traffic should always pass through a same l4-l7 device else the session will break(Symmetric PBR). Also if an L4-L7 device fails the flows should shift to new L4-l7 without affecting the other flows(resilient hashing) etc. All these features are very well designed and time tested with Cisco Nexus DC.

• Scalability – A data center switch fabric must be able to grow quickly as subscriber demand rises. This requires foresight. For smaller Edge DCs, you can often start with vertical scaling (scale-up) by upgrading to higher-capacity switches — for example, using a vPC pair or an ACI Remote Leaf architecture. However, if you anticipate sustained growth in the region, you’ll need to think about horizontal scaling (scale-out). A spine–leaf fabric is ideal here, as it supports incremental expansion and prevents bottlenecks. Cisco provides strong options for both approaches:

Vertical Scaling (Scale-Up): vPC pair as Remote DC extension, or Remote Leaf designs.

Horizontal Scaling (Scale-Out): Smaller PODs in a Multi-Pod architecture. Spine–leaf fabric not only offers ample headroom for expansion but also ensures cleaner, more manageable operations, with central visibility through a single controller.

The similar options are present in CISCO NXOS VxLAN EVPN type of fabric.

• Resilience & Reliability – Streaming downtime is not acceptable; redundancy and fault tolerance are mandatory, and its needless to say that at how many levels these days the DC fabric ensures resiliency. From the vPC that provides redundancy at TOR level to the redundant links to the spine and to the redundant spines and many more.
• Centralized Management – It is very important that all these remote edge Data Centers are fully automated and controller including day0 . Technologies like remote leaf for remote edge DCs suitable for these kind of deployments.
• Some more subtle but useful feature can be 32 active members in port channel which Cisco Nexus Supports.
• Also, did you hear about DLB(dynamic load balancing) ? Port channel hashing has been always prone to polarization, this is solved now with a very intelligent algorithm which also considers link utilization not just src/dst mac/ip etc for hashing. DLB is also very useful for high capacity media streaming POPs.

• Cisco has also come up with a new feature which is going to enhance in upcoming releases called Remote leaf resiliency(RLR) in ACI. This feature makes the remote sites independent of the central DC. This is huge enhancement and provides alot of operational benefits.

Conclusion

While CDN is one of the most common use cases for Edge Data Centers, it’s far from the only one. Multiple market drivers are accelerating Edge rollouts, and we’re bound to see them emerge more frequently. At the core of this trend lies the explosive growth in data traffic, which has doubled in just 3–5 years. To deliver a seamless user experience, it makes far more sense to process and manage that data closer to the customer.

As new use cases continue to expand — from media streaming and online gaming to AI inference, smart cities, and beyond — the future of Edge computing promises to be both exciting and demanding. Keeping pace will be challenging, but also highly rewarding. Businesses that invest in building and operating efficient Edge Data Center ecosystems will be far better positioned to win in this rapidly evolving digital landscape.