Transcription

Master thesis: Data centre design standards and best practices forpublic research HPC centresSeventh Annual Workshop for the Energy Efficiency HPC Working GroupLadina Gilly, CSCSNovember 13, 2016

Agenda Motivation Research questions Interviewees FindingsInsert Footer2

Motivation and goals Motivation stems from personal experience ofchallenges when planning new CSCS datacentre HPC requirements v. design standardsLife cycles of buildings and ITFast changing technology developmentsOptimising CAPEX and OPEXGoals: Provide overview of relevant design standards Examine the drivers that differentrequirements Provide future managers of HPC data centreconstruction projects with a useful startingpoint.Insert Footer3

Research questions To what extent are existing data centre design standards known and appliedwithin public research HPC data centre? Which design issues, common to public research HPC data centres, do thestandards not cover? How do public research HPC data centres define design criteria where thererequirements are not covered by existing standards? What approaches to future-proofing do public research HPC data centres apply? Is there sufficient similarity in the applied approaches to allow the definition of astandard for the design of public research HPC data centres?Insert Footer4

IntervieweesInsert Footer5

Familiarity with and application of data centre design standards Standards surveyedFamiliarity with data centre standards Uptime Institute Data Center TierClassification and Performance Standard ANSI/TIA-942 TelecommunicationsInfrastructure Standard for Data Centers ANSI/BICSI 002-2014, Data Center Design &Implementation Best Practices EN 50600 Information Technology – Datacentre facilities and infrastructures The American Society for Heating,Refrigeration and Air-Conditioning Engineers(ASHRAE) Datacom SeriesUptime InstituteASHRAEEN 50600BISCITIA0510Series 1Insert Footer61520

Where standards do not cover HPC requirementsAttributeData centreindustry1 – 15kWPublic research HPC sites insampleMin. 20 kWMax. 100 kWRaised floorheightRaised floorratingEquipment onUPSUp to 1.1 mMin. 0.4 mMax. 6 mUp to 1220 kg/m2Min. 980 kg/m2Max. 3410 kg/m2AllCoolingtechnologyAirTime to buildUp to 30 months13 sites: critical equipment only6 sites: all equipment1 site: no equipment16 sites: liquid16 sites: hybrid14 sites: airMin. 24 monthsMax. 180 monthsLoad per cabinetInsert Footer Caparison of HPC site attributes and thoseobserved by the Uptime Institute in theenterprise data centre industry show thatthese are the areas where the two differmarkedly: 7Load per cabinetRaised floor heightRaised floor ratingEquipment on UPSCooling technologies in useTime to build

Criteria not covered by standards and challenging to define– rep. 10Insert Footer8

Future-proofing challenges and strategies Design infrastructure to accommodate growth Building envelope – big vs. modular Plan for the future – know your business plan, your assets and your technology Diversity of cooling technologies - different technologies and temperatures Invest in a raised floor with high specificationsInsert Footer9

Scope for definition of a design standard for HPC data centres?Definition of a standard:British Standards Institution: “ . An agreedway of doing something. Standards aredistilled wisdom of people with expertise intheir subject matter and who know the needsof the organizations they represent . Thepoint of a standard is to provide a reliablebasis for people to share the sameexpectations about a product or service.”ISO: “ . a document that providesrequirements, specifications, guidelines orcharacteristics that can be used to ensure thatmaterials, products, processes and servicesare fit for their purpose.”Mission of HPC sites: accelerate scientific discovery tackle issues of societal and scientificimportance push the boundaries of discovery andengineering ensure the competitivity of a nationVs.First of a kind supercomputersLeading edge technologyHPC sites are frequently the firstto gain experience with newsystemsStandards are established based onexperience.Insert Footer10

Best practices – key topics Management topics Tendering processes Building envelope Raised floor Electrical infrastructure Cooling Fire protection Measuring and monitoring Once in operationInsert Footer11

Conclusion Although standards do not cover allrequirements of HPC they are nonethelessuseful as a starting and reference point.Defining the design criteria for an HPCdata centre is a complex task thatrequires a combination of engineeringexpertise, great curiosity and educatedguesses about future IT developmentsto allow the derivation of a clear designrequirement for the A&E partner whilstmaintaining maximum flexibility toadapt to future changes. Due to the nature of HPC, the definition of astandard is not possible. A collection of best practices applied in thepeer community can however provide a goodknowledge base for the community to workwith and exchange and build on. Thecollection compiled during this work is by nomeans exhaustive.Insert Footer21

Thank you for your attention and thanks again to all intervieweesfor contributing their time, experience and knowledge.Want to know more? You can download the thesis at the followinglink: http://www.cscs.ch/publications/technical reports/index.html