High-per­for­mance com­pu­ting (HPC) is now indis­pensable in many sci­en­ti­fic fields, such as phar­maceu­ti­cal rese­arch and cli­mate rese­arch. Howe­ver, data cen­tres equip­ped with highly energy-inten­sive super­com­pu­ters gene­rate signi­fi­cant CO₂ emis­si­ons. ‘Energy-opti­mi­sed super­com­pu­ter net­works through the use of wind energy’ (ESN4NW): this title refers to a new, nati­on­wide col­la­bo­ra­tive pro­ject led by the SICP (Soft­ware Inno­va­tion Cam­pus Pader­born) at the Uni­ver­sity of Pader­born. In col­la­bo­ra­tion with West­fa­len­WIND IT, a SICP mem­ber com­pany, and its brand wind­CORES, the idea arose to deve­lop a new HPC infra­struc­ture based on a sus­tainable con­cept. The Fede­ral Minis­try of Edu­ca­tion and Rese­arch (BMBF) is now fun­ding the pro­ject with around 2.5 mil­lion euros over the next three years. The con­sor­tium also includes the Uni­ver­sity of Pas­sau, the Fraun­ho­fer Insti­tute for Relia­bi­lity and Micro­in­te­gra­tion (IZM), and the com­pa­nies Aix­pert­Soft GmbH, Rit­tal, Atos Ger­many and Zattoo. 

Pro­vi­ding cli­mate-neu­tral power to data centres

Tog­e­ther, the pro­ject part­ners are rese­ar­ching and demons­t­ra­ting the poten­tial of so-cal­led ‘wind­CORES’. These are wind tur­bi­nes that house data cen­tres within their towers and sup­ply them with on-site wind-gene­ra­ted elec­tri­city in a vir­tually cli­mate-neu­tral man­ner. The aim of the pro­ject is to deve­lop the infra­struc­ture and ope­ra­tio­nal manage­ment of an HPC clus­ter within seve­ral wind tur­bi­nes. The direct, locally available rene­wa­ble energy is to be inte­gra­ted into the ope­ra­tio­nal manage­ment sys­tem in order to make maxi­mum use of it. In addi­tion, the waste heat gene­ra­ted is also to be taken into account as a limi­ting factor. 

To demons­trate this, an alter­na­tive archi­tec­tu­ral approach – one that goes bey­ond the cur­rent state of the art – is to be pur­sued and imple­men­ted. The pro­ject part­ners are deve­lo­ping an infra­struc­ture which, through inten­sive data uti­li­sa­tion and arti­fi­cial intel­li­gence (AI) methods, con­ti­nuously adjusts the com­pu­ting power of the decen­tra­li­sed HPC clus­ter via an ope­ra­tio­nal plat­form. The clus­ter is spread across seve­ral wind tur­bi­nes and must make opti­mum use of the fluc­tua­ting energy avai­la­bi­lity. Fur­ther­more, the pro­ject is inves­ti­ga­ting the impact of dif­fe­rent archi­tec­tures and ope­ra­tio­nal stra­te­gies on the sus­taina­bi­lity assess­ment of the tar­get sys­tem. AI-sup­ported ope­ra­tio­nal manage­ment requi­res beha­viou­ral and pre­dic­tive models for the infra­struc­ture and the local energy grid, which are not yet available. “The sci­en­ti­fic and tech­ni­cal chal­lenges of the pro­ject lie, amongst other things, in the inten­sive coll­ec­tion and uti­li­sa­tion of data (energy grid & infra­struc­ture), in the inte­gra­tion of arti­fi­cial intel­li­gence into con­trol loops, and in the sys­te­ma­tic con­so­li­da­tion of all models using a digi­tal twin,” explains Dr Gun­nar Scho­ma­ker, Mana­ger at SICP. “The focus is on pre­dic­tions regar­ding wea­ther-depen­dent local energy avai­la­bi­lity, as well as pre­cise beha­viou­ral models of all inter­ac­ting sys­tems in terms of their power con­sump­tion and heat dis­si­pa­tion,” Scho­ma­ker continues. 

An inno­va­tive con­cept with intel­li­gent circuitry

A distinc­tive fea­ture and inno­va­tion of the pro­ject is that the tower of a wind tur­bine is incor­po­ra­ted as a poten­tial heat sink when no heat sink is available or when the exis­ting one does not curr­ently require the waste heat. The tower can then absorb the waste heat from the IT sys­tems and cool it down. This func­tion, tog­e­ther with energy avai­la­bi­lity, is to be used as a gui­ding prin­ci­ple for the ope­ra­tion of the HPC clus­ter. In prac­ti­cal terms, this means that the sys­tem only runs when rene­wa­ble energy is available and waste heat can be dis­si­pa­ted in a vir­tually car­bon-neu­tral man­ner. Initi­ally, the pro­ject will focus on tech­ni­cal and eco­no­mic fea­si­bi­lity, the func­tional and edu­ca­tio­nal demons­tra­tion of ope­ra­tio­nal manage­ment, and the sus­taina­bi­lity assess­ment of the sys­tem over its exten­ded life cycle. Howe­ver, the inten­ded results are inten­ded to be trans­fera­ble not only to wind tur­bi­nes but also to other HPC sys­tems with waste heat gene­ra­tors and waste heat users. 

“The pro­ject spe­ci­fi­cally addres­ses the chal­lenges posed by the energy tran­si­tion and digi­ta­li­sa­tion. We want to demons­trate that the rising energy demand dri­ven by digi­ta­li­sa­tion does not pre­sent an insur­moun­ta­ble obs­ta­cle to grea­ter sus­taina­bi­lity, and that these growth requi­re­ments can be met fle­xi­bly – both in terms of time and loca­tion – through rene­wa­ble energy,” says Dr Fiete Dub­berke, Mana­ging Direc­tor of West­fa­len­WIND IT, describ­ing the project’s objec­ti­ves. “In doing so, we are asking our­sel­ves the fol­lo­wing ques­ti­ons: How can we manage waste heat from a wide variety of indus­trial sec­tors – in other words, how can we uti­lise it? Which sys­tems com­pri­sing gene­ra­tors and con­su­mers should be sen­si­bly lin­ked, and which tasks can be hand­led by arti­fi­cial intel­li­gence methods?”, explains Scho­ma­ker. The “wind­CORES” con­cept, which won the “Ger­man Data Centre Award”, has alre­ady demons­tra­ted that inte­gra­ting a wind tur­bine with a data centre can be imple­men­ted suc­cessfully from an eco­no­mic perspective.