Through plug-and-play energy optimisation for mid and small tier data centres, Interact is an extremely welcome development in the sector’s journey towards sustainability
The tool helping enterprise data centres net quick green wins
Thu 4 Feb 2021
“This tool will benefit almost every mid and small tier data centre,” says Richard Kenny, director of strategy at Interact, a newly-launched application enabling data centre operators to select the most energy efficient hardware for their facilities.
By easily identifying the hardware and configurations most suited to individual performance requirements, Interact is attracting attention from enterprise data centre operators looking to save energy costs, server room real estate and progress a green agenda.
Connecting the dots
Interact is the culmination of a two-year collaboration between the University of East London and refurbished hardware seller Techbuyer, funded by Innovate UK.
The partnership’s goals when conceived in 2017 were to leverage developments in data science and advanced modelling to create a tool capable of identifying the most energy efficient hardware for any data centre scenario, and substantiate the principles of the circular economy.
Through newly-published research in prestigious computer science journal IEEE, the team behind Interact has proven both the energy-saving credentials of the tool it built and the surprising performance merits of refurbished hardware.
New vs old
According to the paper, refurbished and new hardware offer comparable and even identical performance when correctly configured. The research is backed up with wide-ranging case studies that underscore the financial case for choosing used hardware over the latest tech.
The paper’s scrutinisation of a floundering Moore’s Law, referencing typical server utilisation rates, also shows how newer hardware is actually yielding diminishing efficiency returns.
While the latest data centre kit does boast better performance at full tilt (90% utilisation), the average customer hovering between 25 and 30 percent utilisation actually sees a drop in performance by opting for out-of-the-box hardware.
“Unless you’re running a high performance compute-machine, and particularly complex AI, these efficiencies are not actually real. They’re conceptual,” explains Kenny.
Despite having an obvious interest in extolling the virtues of used hardware, Kenny said Interact is vendor and condition agnostic: “While the IEEE research shows refurb does yield much better benefits on a cost-horizon, whether it’s 10 years old or out the box, Interact finds the right solution.’
Interact’s small but mighty team is composed of firstly Kenny (also Group IT Director at Techbuyer) and Astrid Wynne, sustainability and communications manager at Techbuyer. They bring decades’ experience in the data centre hardware sector, and crucially Techbuyer’s vast archive of hardware product data Interact was largely modelled on.
Lead developer and researcher, Nour Rteil, under the guidance of her academic lead Rabih Bashroush of the University of East London, did most of the heavy lifting. Aside from co-authoring the IEEE paper, the American University of Beirut graduate and former Ericsson intern built Interact from scratch, trained its machine learning models, and conducted the preliminary theoretical research and subsequent hardware benchmarking.
The latter presented the project’s biggest challenge, she admitted.
“We were worried the benchmarking experiments wouldn’t produce the results that we were seeing on paper.” Benchmarking, testing real performance at different loads, is complex and involves a lot of trial and error. There was not only the chance that real-world hardware performance would challenge Nour’s predictions, but the possibility of incorrect benchmarking producing false positives.
Techbuyer’s configuration expertise supported Nour’s benchmarking setup to ensure it robustly validated the proposed theory and methodology, helping her install the benchmarking tool on various servers in a multitude of configurations.
Robust the methodology certainly was. Benchmarking eventually validated the theoretical research in a range of environmental conditions. Special mention to the 2019 edition of Data Centre World Frankfurt, where successful benchmarking on a rack at the show’s onsite build (one unseen by the team until the day of the event) further evidenced Interact’s far-reaching applicability.
“We kind of went as far as we possibly could to say this isn’t something we ‘hoped’ would work. We picked someone else’s rack at random in front of 1000s of people,” explained Kenny.
Kenny says Interact is best suited to 15 to 55 rack mid-tier data centres (hyperscalers use their own fine-tuned configuration tools), though as small as 6-server data centres can save tens of thousands in annual costs by implementing its suggestions. In an edge deployment, where space is at an ultra-premium, Interact’s ability to rapidly consolidate server footprint is also pretty compelling.
When recently tested on a 35-rack enterprise/multi-tenant data centre, fitted with hardware from a range of vendors, Interact showed how the facility could achieve the same performance by consolidating its 770 servers down to just 160.
In theory, over five years this translates to energy savings of 4.5 million Kilowatts (representing half a million pounds in costs) and a 1100 tonnes reduction in Co2 output — in practice, colocation operators can’t start binning customers’ servers.
Nevertheless by offering plug-and-play energy optimisation for mid and small-tier data centres, Interact is an extremely promising development in the data centre sector’s journey towards sustainability.