Interact: low carbon, circular, energy efficient data centre IT hardware optimisation
Q. What exactly is the product / innovation?
Interact’s first of its kind machine learning algorithm assesses the energy draw and performance of a data centre’s servers and helps to optimise energy usage, associated carbon, supply chain emissions and cost. Specifically, the tool carries out component and rack level analysis of data centre IT hardware, then calculates the optimum number and configuration of replacement servers, and generates a clear report on best solutions for cost and for energy efficiency over time.
Interact is a start-up based in Harrogate, UK that launched at the end of 2020. Providing vendor neutral, evidence-based reports that demonstrate the benefits of refurbished and remanufactured servers hastens the arrival of the circular economy in enterprise data centres. It optimises energy, calculates Carbon Usage Effectiveness, and reduces supply chain emissions.
Q. How does it break conventional ideas or processes in its field?
Sustainable IT solutions specialist, Techbuyer researched the materials content of servers using information from the JRC committee informing the EU Ecodesign directive and Dell carbon footprint information. These sources revealed that servers contain more than one metric ton of CO2e to manufacture and also use a significant number of the 30 Critical Raw Materials identified by the EU as in low or politically unstable supply. With recycling technologies unable to recover anything near 100% of these materials (and none of the supply chain carbon), product life extension in the form of remanufacturing, reuse and upgrades is the best solution.
The market operated under the assumption that new machines were more sustainable by default because of the high rate of performance gains with each successive generation (known generally as Moore’s Law). However, with the slow-down in Moore’s Law, Techbuyer saw an opportunity to carry out scientific research to call this assumption into question.
Contributing to new research:
Techbuyer IT Director Rich Kenny and Sustainability Lead Astrid Wynne contacted leading academic research into energy usage in relation to compute power with Rabih Bashroush of the University of East London (UEL). Together they developed a 2-year Innovative UK Knowledge Transfer Project (KTP) to establish the performance and energy draw of different generations of server and upgrade configurations. Expert developer and researcher Nour Rteil joined the team to carry out thousands of hours of benchmarking and analysis on multiple generations, brands and configurations of servers.
The results of the research (recently published in the IEEE Transactions on Sustainable Computing) proved that not only were refurbished machines equivalent to their new alternatives but also an immediate past generation could outperform the latest with the correct upgrades. This lead to cost analysis of refreshing machines over five years old with various later generations. Return on investment was much shorter even with electricity usage factored in with certain older generations.
This information led to the development of the Interact tool, which used 400 pre-configured servers and 1400 pre-configured server records to assess performance in comparison with energy draw in data centres. The tool identifies the worst performing slice of a server estate and then suggests the best hardware to replace this with optimum energy efficiency gains and cost reductions, using machine learning to improve analysis as it goes.
Interact team members were all part of the original research project. This research has turned traditional ways of thinking on their head and made the case for increased circularity. Whereas in the past IT managers believed that refreshing the IT estate with new servers would be more efficient by default, these experiments proved that this is no longer the case. Interact provides unique server recommendations to capitalise on these findings.
Q. How does the product go beyond marginal improvements of something that already exists?
Interact rethinks the way that people manage their IT infrastructure. It is the first tool of its kind to balance carbon avoidance with the use of refurbished severs (therefore reducing the necessity for new manufacture) with carbon avoidance through energy efficiency measures. In short, it enables data centre managers to reuse hardware without losing out on the energy bill.
As different regions have different proportions of green energy in the grid, the tool includes the facility to change the carbon output associated with energy usage, which in turn affects the proportion of embodied energy performance. The rational for this is that servers are in the main made and assembled in areas where the energy mix still includes a lot of fossil fuel generation.
Interact is also non-intrusive. No complex infrastructure is necessary unlike others on the market – users simply tell us what they have. The tool then carries out component and rack level analysis of existing infrastructure and generates vendor-neutral solutions. Finally, users receive a report with cost analysis and energy usage projections. These accurate figures help professionals to make the best decisions for their business.
Q. How do customers benefit from it?
Interact generates clear sustainable and business benefits for businesses. From our case studies we have provided savings of up to 53% on energy use, 48% on C02 reduction (scope 2), 80% reduction in scope 3 C02, and operational savings in excess of £1.1million pounds or 42% of estate cost.
Using Interact’s first software tool, Techbuyer reduced 6 of their newer servers down to just 3 refurbished machines from a different brand. This enabled 17,429 kWh savings in energy draw, 4,460 kg CO2e reduction on the energy spend, 2,735 kg CO2e in supply chain carbon and £9081 p.a. in cost savings. The company was able to share best practice in a case study for a local climate change commission.
Within the larger Techbuyer group, Interact training has been given to members of Techbuyer’s sales teams in order to deliver the best energy efficient solutions on their refurbished server solutions. It enabled customers to make the right choices on price, energy efficiency and return on investment over time, generating a solid business case as well as a quantifiable sustainability benefits. In addition, the IEEE paper delivered point by point, academically verified answers on the benefits of refurbished hardware.
Q. What tangible benefits has the organisation seen as a result of the project's implementation?
Interact is currently delivered in the UK and in Germany by partners in the resale and environmental consultancy channels. These include Data Centres 4U in the UK and High Knowledge GMBH, both of which have dedicated service portals on their websites. The Interact team has created user training videos in both English and German as part of the onboarding process of these partner companies. It has also collaborated with the partners to create cross marketing collaterals with dual branding.
The start-up has also recently become a sub-partner on an international research project, Circular Economy in the Data Centre Industry (CEDaCI). CEDaCI brings together academics, industry experts and non-profits to increase circularity in the data centre sector across the UK, France, Germany and the Netherlands.
Thought leadership is a big part of the Interact identity, and all three members of the Interact leadership team regularly write for the industry press and appear on panels and keynote addresses at industry conferences. Most recent examples of this are a keynote address at Data Center Dynamic’s Towards Zero Carbon event, and a policy overview in DCD magazine. Upcoming events include a panel discussion on Circular Economy at Data Center World Frankfurt.
Q. What major challenges were faced during the project and how were they overcome?
The perception that refurbished does not perform as well as new is a widespread belief in the IT industry. Overcoming this took thousands of server experiments, two years of research and an IEEE validated paper. There is also a lack of understanding about the cost of server inefficiency, which we have been helping to raise awareness of through events like Global Reman Day as well as quality research.
On the technical side, one challenge was that there are limitations on upgrades possible in servers, which took extensive research into what is and isn’t possible, experiments, and the expertise of the research partner, Techbuyer. Whilst developing the tool we also had to ensure there was no bias in the server recommendations towards certain vendors or models. For that purpose, we made sure to include as many pre-configured server models from several vendors, and different price ranges, in the database. Having a large dataset of preconfigured servers safeguards the neutrality of the recommendations, which differ based on the specified data centre parameters and workload.
An additional foreseeable challenge is making sure that the pre-trained machine learning model produces accurate estimates for newly released servers that are not included in the training set. For that reason, we plan to continuously train and improve the model whenever a new server gets released and tested (SPECpower_ssj2008 benchmark). This will guarantee accurate results for previous and new server releases, factoring in potential CPU advancements that will affect the servers’ performance and energy consumption.