105886Server Modelling CapabilityClosedCollaborative R&DInnovate UKData centres are the infrastructure on which all modern internet systems (such as the Cloud, 5G, and video on demand) rely. They use huge amounts of power - in the UK they consume 3% of the National Grid, and are growing so quickly that it is estimated they will consume 9% of the entire world's power supply by 2030. The industry has invested heavily in hardware improvements, but there has been very little research into improving efficiency though software. Edgetic has developed a software resource management system for data centres that can reduce power consumption by up to 20%, as tested in real data centres. To do this, computing server behaviour needs to be modelled to great accuracy - this lets us understand how different servers respond to differing software loads (for example, how much power does it consume when running a typical workload? How much heat does it produce? etc) There are many problems with existing modelling - it is slow, is performed on basic hardware, and does not consider many environmental factors. This project will create the most advanced server modelling facility in the world. The project involves Edgetic designing and building a new testing environment that features an advanced wind tunnel with software-controllable components (including humidity controls, heaters, variable fan speeds, etc.). This is a step-change over existing designs - which feature fixed fans and heating, with fixed size testing enclosures - and will be the most advanced server modelling wind tunnel in the world, allowing over 100 different parameters of a server to be modelled far quicker than anything that currently exits. The project will also develop advanced machine-learning software which will be used to control the wind-tunnel and also analyse the data it produces. This software will automatically "balance" the wind tunnel's components - for example, ensuring that testing temperatures remain the same by automatically increasing heating when fan speeds change to increase air flow. The software will also interpolate testing data - automatically "joining the dots" between different tests, and learning and applying patterns from previous models. This will allow tests to be performed with greater speed and accuracy than anything currently seen. The result will be accurate, detailed server models that can be quickly produced - and used to drive resource management software that will greatly reduce data centre power consumption, helping the environment while also saving businesses substantial amounts of money.