With AI-fuelled HPC in big demand, Paul Mellon, Operations Director, Stellium Datacenters, tells us why Open Compute’s OCP-Ready certification is the key to keeping colocation data centres and their users in step.
Not too long ago, a data centre – albeit a fairly small one – might be considered to be a cluster of 50 to 100 racks with a combined IT load of 100kW. A single rack can now accommodate 100kW in normal chilled water cooling and this scales to 250kW with immersive cooling.
These developments represent giant steps forward in terms of efficiency and have stemmed from necessity as IT users demand more and more compute resources. This will only accelerate as AI-based applications take hold.
However, achieving such breath-taking innovation in ever-decreasing timescales would be impossible to replicate in a normal business environment. It could and can only come from research and development on a mass scale, underpinned by industry standardisation.
The Open Compute Foundation (OCP) has remained at the forefront of such industry initiatives for more than 10 years to short circuit the development and standardisation process, from decades to years and sometimes months.
As a result, high power density and cooling is now readily available at all levels in the data centre community which has in turn allowed the migration of millions of IT business environments to the cloud – or data centres. The benefit of these migrations is not just raw efficiency in terms for lower power usage. There are substantial benefits in terms of flexibility of how organisations choose to manage their services and growth as well as fundamentally robust SLA’s guaranteeing 99.98% service availability.
No surprises
The silver bullet here is certainty. Data centre operators and their users – solutions architects a primary example – both require certainty of outcome. With this in mind, the OCP Ready certification program for colocation data centres was conceived and launched around five years ago.
Its goal is to help systems designers and integrators identify colocation data centre facilities where OCP IT equipment can be deployed without complications. Because the program’s standards are open-source (like some software), operators will know what a given rack needs to do in terms of size, capability and power before it arrives onsite, and ensure the design and layout is ready to support this class of HPC equipment. The net result: faster speed to market and reduced costs.
Efficiency, impact, openness, scale and sustainability represent the core principles of living in the OCP community. However, it’s quite a challenge for data centre operators to join the OCP and therefore be eligible to undergo the detailed OCP-Ready self-certification process. It entails working with the OCP to achieve compliance against their rigorous criteria for power, cooling, IT technical space layout and design, facility management and control and facility operations. Thereafter, demonstrating an ongoing commitment to continuous improvement and sharing. The certification is reviewed every two years to track development against the core principles and PUE.
At the outset, it involves presenting your data centre technically and environmentally to your technical peers including a Q&A session – we had to demonstrate that we lived their principles in our daily operations. This is key as it just wouldn’t work if these remained on the wall as a set of rules. We also had to demonstrate our commitment to continuous development in terms of ISO14001 – Environmental Management, our CCA tariff relief programme and to lowering our PUE from its current level of 1.24.
Raising the bar
The majority of Open Compute hardware is deployed as a fully populated rack. OCP-Ready data centre facilities must be able to easily accommodate racks of these weights and dimensions and be able to deploy multiple racks of such equipment at scale. As a minimum, the rack sizes can weigh 500kGs to 1500kGs and 47U in height. The workloads can be anything from 6.6kW to 36kW and beyond.
In addition, the density of compute to meet scale demands and efficiency goals raises the bar for power and cooling specifications for racks. Cooling of the racks will require potentially a range of solutions – immersive, chilled water/air, chip cooling. For the higher densities the expectation is for PUE to be sub 1.1.
OCP-Ready also fits neatly alongside other global standards within the industry such as BREAM, LEED, ISO and Uptime while complementing The European Code of Conduct for Data Centres (EU DC CoC) which Stellium also supports.
It just makes sense at some point that all these design and operational standards converge to rationalise the processes. All likeminded organisations within OCP will find they have significant commitment to BREAM, LEED, ISO and Uptime. It remains for our members to lobby these other organisations to inspire cooperation and sharing of our common values and mission.
In summary, with the growing challenge of supporting HPC AI-based applications and services with fit for purpose data centre infrastructure, the OCP-Ready certification program offers colocation operators and their customers much needed clarity and certainty. They can be assured that the global organisation behind it today has over 8,000 engineers and members such as Arm, Meta, Google, HPE, Inspur Systems, Intel, Microsoft, NVIDIA and others. The mission now embraces a much wider vision and goes well beyond optimising computing, storage and network efficiency.
Click below to share this article