Articles

Even before the multi-cloud era the holy grail for big IT was the ability to move workloads across different infrastructure platforms. Today, all companies, from hyperscale cloud providers, big SAAS and IAAS providers to enterprise players strive to make applications, dev and ops more dynamic – i.e., host workloads in the best place and move it as necessary.

The goal is to literally abstract the workload from the infrastructure. The open source orchestration revolution of Kubernetes has propelled much of the progress in this arena, e.g., by enabling the world of flexible Cloud SLAs. However, it is not the only catalyst and different approaches are being tried with firms throwing vast amounts of resources and R&D at the problem. For example, some hyperscalers say they can dynamically move workloads to locations where renewable energy is available to power their data centres. But creating fully dynamic environments where workloads move automatically is a highly complex undertaking and no-one has so far cracked this conundrum at scale. The objective is clear, but even as strides are made, how data centre power can match this dynamism needs consideration.

In almost all data centres, power is literally the life blood of the operation. However, the power topology is fixed at the design stage and is uniformly delivered throughout the facility. In an age of dynamic environments, the question arises whether power can be made more adaptable in order that data centre power chains can be responsive to accommodate multiple IT SLAs?

As a general rule of thumb, data centres don’t tend to be operated utilising above 70 percent of their total power. In addition to this headroom maintained for reasons of, e.g., redundancy, some data centre power can also become stranded, or trapped as a result of IT deployments when equipment is added, shifted or replaced in the racks. This gives rise to a second set of questions; is possible to capture and divert this inherent data centre power to where it is needed?

Can this be done without adversely affecting redundancy? And is it possible to ensure that flexibility does not come at a price of increased risk of disaster – i.e., to ensure disaster avoidance when the unthinkable happens and grid power is lost?

In short, is it possible to use automation as a means to direct power to the most critical applications and workloads when outage disaster strikes? Can a data centre power system be designed incorporating automated dynamic provisioning where power for environments is regulated to meet SLAs using adaptable redundancy?

The answer is yes. The answer is ARP.
Adaptable Redundant Power (ARP) topologies can deliver exactly this. And more. It is based on four operating modes, Adaptable Redundancy; Inherent Redundancy; Adaptable Inherent Redundancy; Disaster Avoidance.

Adaptable redundancy means companies can achieve the goal of automatically powering applications that are dynamically provisioned as containers even where applications have different SLAs in the same data centre. It does this by overcoming the power system single SLA which inevitably means the power system is either under provisioned or over provisioned. Adaptable redundancy is the ability to dynamically modulate power system availability to align with variable IT SLAs. It provides variable redundancies, 2N, N, N+1 and N to different cabinets, rows or data halls as needed. Inherent Redundancy is the ability to access stranded power that already exists in the power chain and to use it to provide virtual redundancy. For example, in a data centre comprised of six 1MW halls, the power drawn will typically average around 60% of the power available and in almost every case never peak above 80%. Inherent redundancy enables that stranded accumulated power capacity to be directed as needed with no
physical modifications required to the power system.

When everything is operating as normal in a data centre all applications are equal. That is until there is an outage. AR disaster avoidance means power can be automatically directed to the most critical workloads in the event of extended power loss. In any outage a partial power system failure resulting in reduced capacity can cause a total data centre outage. Using ARP, disaster is avoided as it prevents a cascade data centre failure by preserving power to the most critical applications.

Be adaptable and automated
The pressures on data centre operators to deliver increased sustainability by making better use of existing resources, to be more responsive and flexible with power provision to match workloads moving around while maintaining current levels of resilience and reliability are real. Adaptable power provision is already widely used in other sectors such as critical processes and energy. The key is automation and it is time for the data centre sector to be pragmatic and look at the benefits of Adaptable Redundant Power.

Article first published in DCS InnoVision Digital Magazine - September 2021.