How Netflix’s cloud journey reshaped modern computing

Flexibility, modularity and scalability – while the buzzwords bandied around in the discourse on digital technology can get rather boring, tracing back how state-of-the-art technologies were combined to deliver these qualities never ceases to fascinate.

 

Behind every major shift in modern computing there is a story of incremental problem-solving and trial-and-error far less glamorous than the hyped language that surrounds it. And few stories illustrate this interplay of necessity and innovation better than Netflix’s.

 

Most people are familiar with the victory of Netflix – an innovative start-up at the time – over DVD hiring giant Blockbuster. What is less known, though is how the company was doggedly putting together the technological puzzle pieces that underpin much of today’s cloud-native ecosystem.

 

Netflix’s innovations weren’t created for the sake of it, but to address tangible operational bottlenecks. When its on-premises servers couldn’t cope with the steeply growing number of subscribers, Netflix embarked on what would become an eight-year-long migration project to the cloud, becoming completely cloud-native in January 2016.

The shift was also influenced by a pivotal three-day outage caused by a faulty hardware component in Netflix’s monolithic application for its DVD-by-mail service, which dramatically exposed the vulnerability of monolithic architectures that were still the industry standard in the middle of the noughties.

 

To make the most of the elasticity cloud services offered, Netflix gradually broke down its DVD rental application into microservices, which also enabled the scaling, debugging and updating of certain services why the others remained intact.

 

Although there was no central company strategy to bundle these microservices into containers, individual Netflix engineering teams – having realised the flexibility that decoupling applications from the underlying infrastructure allows – organically transitioned to them.

 

By 2015, Netflix had developed and deployed its own proprietary container management platform too.

At that time, Kubernetes (K8s) – today’s dominant container orchestration tool – was still in its infancy. Yet Netflix was already launching millions of containers weekly, pioneering approaches that others would only adopt years later.

Fast forward ten years and the technologies that Netflix either pioneered or was an early adopter of are accessible to most companies in one form or another.

 

However, whether an individual business can actually take full advantage of these tools remains an entirely different question.

 

Making the most of the cloud

 

In this Russian doll of cloud technologies, the smallest unit is the microservice – each providing a discrete piece of application functionality.  

 

Containers group these microservices and wrap them with everything they need to run consistently and securely.

Container orchestration sits on the outside as the largest doll, co-ordinating and automating everything happening inside.

 

Without this top layer of automation, IT teams must carry out all tasks manually throughout a container’s lifecycle from deployment to monitoring, networking, scaling and recovery.

 

That includes starting and stopping containers, routing traffic between services, performing health checks, restarting failed containers and moving workloads.

 

While manageable for a small number of containers, this quickly becomes untenable when the numbers reach the 20s, especially where complexity is compounded by a multi-cloud environment.

 

Unlike some other big tech spaces, cloud services are highly competitive with each provider having their own strengths.

 

For a global infrastructure and the widest array of services, Amazon’s AWS is the go-to. Azure is the natural fit for organisations deeply embedded in the Microsoft ecosystem, and Google Cloud attracts businesses seeking analytics and ML capabilities.

 

An orchestration platform is essential for stitching these different environments together, enabling companies to pick and choose the best services while maintaining consistency across clouds.

 

The same applies to hybrid cloud environments, where legacy in-house servers also remain part of the equation.

According to the latest cloud computing statistics commissioned by IT and communications company AAG, UK business increasingly seek multi-cloud solutions, with 89 per cent of respondents using multi-cloud, most typically a blend of public cloud, private cloud and on-premises infrastructure. This rising complexity makes orchestration indispensable.

 

Kubernetes’s claim to fame and market dominance

 

Initially a niche Google project launched in 2014, by 2017 Kubernetes was the de facto standard for container management, when all major cloud providers announced fully managed Kubernetes offerings. Its rapid uptake made it the common language of modern software development – though not an easy one to learn.

 

Kubernetes’s success lies in how it juggles workloads across pods – the basic units separating applications inside containers from underlying physical and virtual servers called nodes in order to achieve portability.

 

This portability is the feature that empowers Kubernetes to scale down pods or even entire nodes if traffic falls at night or spin off new ones if demand spikes. Also, if a node with all its CPU and RAM resources goes down, Kubernetes can reschedule its pods to healthy ones.

 

The beauty of Kubernetes for cyber-security professionals is that security settings bundled into pods ensure that constant reshuffling does not create vulnerabilities.

 

Kubernetes also distributes network traffic efficiently across pods, ensuring optimal resource utilisation and facilitating seamless scalability.

 

But even this simplified explanation reveals Kubernetes’s inherent complexity.

 

While it’s used by over 50,000 companies globally and enterprise adoption rates are over 60 per cent, developers often criticise its extensive and heavy jargon and high operational overhead, causing teams to spend more time on managing clusters of nodes than developing new software features.

 

Avoiding overkill

 

As always, the “no one size fits all” shibboleth holds painfully true. Large organisations with big IT teams, advanced operational maturity and microservice-heavy architectures may find Kubernetes indispensable.

 

Others, particularly SMEs, however, may feel that they have traded simple problems for far more complicated ones.

Fortunately, today’s orchestration landscape is much broader than it was during Kubernetes’s early rise. There are several alternatives that aim to strike a better balance between control and usability, providing up to 90 per cent of Kubernetes’s flexibility with as little as 20 per cent of the complexity.

 

Cloud providers now also offer serverless container platforms where they handle the servers, virtual machines and underlying clusters entirely, allowing developers to deploy applications with a single command.

 

Internal development platforms (IDPs) built in-house by an organisation’s platform engineering team are also gaining popularity. These are designed to reduce developers’ cognitive load by providing “golden paths” – templates and automated workflows that standardise how apps are developed, deployed and managed.

 

For businesses whose deployment frequency is modest, and orchestration needs are limited, yet simpler options exist. Running containers directly on virtual machines, supported by lightweight orchestration tools – although less sophisticated than Kubernetes – may provide precisely the balance of costs and benefits they are looking for. 

The good news is that while Kubernetes is undoubtedly here to stay, cloud orchestration diversity has grown – delightfully – in the shadow of its dominance.

 

This widening choice means that ordinary businesses can also harness the technologies that made Netflix great without deploying oversized technologies or overwhelming their developer teams.  All they need to do is make the right choices to find the tools that will serve them best.