Kubernetes opened the world of flexible Cloud SLAs to everyone - It is time for power provision to catch up

Cloud native applications at scale have completely changed computing – supply side data centre power infrastructure must not hold it back

Why are application SLAs and data centre power SLAs not aligned? The software containers and Kubernetes open source orchestration revolution has given us the modern cloud industry and changed how enterprise software companies serve their customers.

But at the physical layer too often these services are running on power infrastructure where provision is fixed or inflexible.

Within the cloud environments of the biggest enterprise software providers this raises potential risks including increasing the risk of service downtime and outages and adding fixed cost to the business.

An examination of the history of cloud reveals how the flexibility provided by a software driven revolution have not been met in kind by flexibility at the physical data centre level.

It was back in the early part of the last decade that the promise of almost infinite capacity offered by public, private and hybrid cloud infrastructure build outs spurred some very smart software developers to look for ways to build and distribute applications at scale.

As Brendan Burns, one of the original Kubernetes (k8s) developers describes in the blog,  history of Kubernetes and the community behind it’: ‘Back in 2013 the cloud was a very different place…Netflix was popularizing the idea of immutable infrastructure...The notion of orchestration, and certainly container orchestration existed in a few internet scale companies, but not in cloud and certainly not in the enterprise.”

Kubernetes was launched in June 2014. It is now managed by the Cloud Native Computing Foundation. For many it has become the de facto standard for cloud native container orchestration. The beauty of containers is that they weigh less heavily on VMs, use less memory and CPU and so opened the world of cloud native development and distribution.

For development teams Platform-as-a-Service (PaaS) Kubernetes provides a constantly shifting hardware layer abstraction. In production environments it is the foundation of service flexibility based on an invisible and abstracted hardware layer. The amazing developments in orchestration tools enabled the ability to deliver distributed applications at scale.

A pervasive characteristic of cloud is that applications are provided at different service level agreements.

More flexible cloud, same inflexible power

The Cloud is also physical and will not work without well run data centres in different regions and territories. As native cloud software provision has become dynamic what has been left behind is the notion of dynamic management of the power infrastructure equipment within data centres.

For the providers this points to an infrastructure challenge which sits not with developers and users but with the infrastructure teams within the services providers. For example, some providers choose to operate active-active data centres to provide maximum resilience.

ServiceNow recently announced two new data centre locations in Europe. ServiceNow runs active-active data centres and established “two new pairings for the hosting of customer instances: a UK sovereign pair between London and Newport, and an EU sovereign pair between Amsterdam and Dublin.”

The company believes that this infrastructure development ensures the continued support for customers’ demand for cloud-native technologies, as they leave legacy data centres for new ones and accelerate their digital transformation journeys with increased cloud-adoption.

Elsewhere, Oracle has four cloud regions in the US and Canada, five in EMEA and eight in APAC. The range of cloud services it provides span Cloud Infrastructure Compute, NOSQL Database to Cloud Infrastructure Archive Classic.

Recent statements from Oracle indicate that enterprises demand more than just availability from their cloud infrastructure. Mission-critical workloads also require consistent performance, and the ability to manage, monitor, and modify resources running in the cloud at any time.

A snapshot of Oracle Cloud Infrastructure Services show the different levels of monthly availability offered for separate workloads.

[ninja_tables id="6093"]

The key here is SLAs. Not all applications are equal and not all services are delivered to all markets.

Is Power Infrastructure Stuck?

Service providers are guaranteeing different levels of availability for different loads. Yet at the power infrastructure layer the world is offered a fixed inflexible power system SLA that host a mixture of application SLAs.

Because the services rely on data centre power infrastructure that was fixed at the time of construction this leads to operating environments which are suboptimal except on the rare occasion the application environment is entirely homogenous and has a single fixed SLA. In the vast majority of cases data centres host applications with multiple SLAs. Inevitably this means the physical layer has either under-provisioned the power SLA – which is risky, or has an over-provisioning power SLA – which is wasteful and costly.

A key question for service provider data centre operations – whether in owner managed or commercial data centres – is whether it is possible to align power provisioning with dynamic application provisioning without ripping and replacing entire power chains in the data centre?

Technology developments happen in hardware as well as software, and the latest thinking in power supply throughout the data centre would provision power to a customer’s needs rather risk power under-provisioning or power overkill.

The characteristics of Kubernetes orchestration include:

  • Controlling resource consumption by application or team
  • Evenly spreading application load across a host infrastructure
  • Automatically load balancing requests across the different instances of an application
  • Monitoring resource consumption and resource limits to automatically stop applications from consuming too many resources and restarting the applications again
  • Moving an application instance from one host to another if there is a shortage of resources in a host, or if the host dies
  • Automatically leveraging additional resources made available when a new host is added to the cluster

Until now cloud service providers have been stuck with inflexible power SLAs which make it  impossible to operate efficiently and effectively in any multiple SLA environment.

This is changing with Adaptable Redundant Power – the patent pending solution i3 Solutions Group. It is possible to align the power system SLA and dynamically provision power to particular rows, racks in the data hall with the application SLA.

For example, the same physical infrastructure may host variable application SLAs such as SQL Database: 99.995%; Machine Learning: 99.990%; Analytics: 99.950% DevOps: 99.900%.

ARP now makes it possible to match and dynamically change the power SLA to match the changing application environment.

Power related issues continue to be problematic in terms of loss of service and cost to provision. ARP will not solve these issues entirely, but it will certainly go a long way to assuaging them.