Post on 21-Nov-2014
description
Kajaani Datacenter Case-‐Study
Peter Jenkins System Specialist & Project Manager
CSC – IT Centre for Science
Contents
! About CSC – High Performance Computing
! Drivers – Why a new datacenter? – Why Kajaani?
! The project ! What we built ! Construction and commissioning ! Lessons learned
I’m from there
Who am I?
! Computer Science background ! 6 years at Sun Microsystems ! HPC consultant in Financial Services ! 3 years in Finland working for CSC ! I’m not a datacenter specialist!
– It helps!
My role at CSC
IT • Rapid change • New hardware
challenging datacenters
Facilities • Traditionally very stable • Disruptive innovations in
last 5 years
• Traditionally separate departments must now work together
• I try to bridge the gap
About CSC • 100% owned by ministry of educaGon
– Public sector, Non profit • Major service areas:
– High Performance CompuGng (HPC) – Storage and archive – FUNET – Academic network in Finland – Managed services and IaaS cloud – Server hosGng and co-‐lo
CSC and High Performance Computing
CSC CompuGng Capacity 1989–2011
338
227
146 141
124
95
42 37
35 34 167
Users of CSC resources by discipline 2011 (total 1386 users)
Biosciences
Physics
Nanoscience
Language research
Chemistry
Grid usage
ComputaGonal fluid dynamics
Structural analysis
ComputaGonal drug design
Earth sciences
Other
WHY A NEW DATACENTER?
0
1
2
3
4
5
6
7
8
9
10
2005 2006 2007 2008 2009 2010 2011
Energy GWh
CSCs Datacenter Energy Use 2005 -‐ 2011
DC 2 Infra DC 2 IT
DC 1 Infra DC 1 IT
-‐ €
10 €
20 €
30 €
40 €
50 €
60 €
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011
Elspot prices FI in EUR/MWh
New site drivers
• Capacity limiGng growth – Super computers – Managed hosGng
• Costs increasing – Tax – Energy – Carbon neutral
• New site focused on efficiency and capacity
0
1
2
3
4
5
6
7
8
9
10
2005 2006 2007 2008 2009 2010 2011
Energy GWh
CSCs Datacenter Energy Use 2005 -‐ 2011
DC 2 Infra DC 2 IT
DC 1 Infra DC 1 IT
2011 Energy cost per MW/hour
Espoo Kajaani
VAT Tax Transfer Electricity
WHY KAJAANI?
10/10/2012 18
Where is Kajaani?
national grid
fibre
fibre
Electricity reliability
29
Finnish naGonal grid Renforsin Ranta substaGon
Fingrid connections to Kajaani
Power capabilities
! Within perimeter fence: – National grid connection access to 340 MW – 110 kV / 10 kV main transformer capacity
! Current capacity 240 MW
– Biopower on site
! Green power options – 3 hydro power plants within 3 km
! feeding directly to site.
! Diverse power supply = reliable power
Related datacenter sites
Facebook: Luleå 120 MW Free air cooling
CSC: Kajaani 1.4 MW Free air cooling 0.9 MW Water cooling PUE design
Google: Hamina ??? MW Sea water cooling
CSC: Espoo 1.6 MW ConvenGonal PUE 1.4 & 1.8
Government support
! Google Hamina = wakeup call – Unused assets ideal for fast growing industry – Jobs, skills, international competitiveness
! Government acted – Regional development money – Extra money to CSC to build a new site
! Site selection: long story short – Initial concept study 2010 – Several former paper mills considered – Kajaani was successful in bidding
THE PROJECT
Approach
! Design goal: multi-MW facility PUE <1.2 ! Leverage features of site
– Matched to business requirements – Avoid redundancy and backup
! Only 100kW UPS from day one – < 5% of load – core network, management, automation – Emphasis on monitoring and rapid recovery
! No generators day one ! Option to add 100% UPS and generators
Approach continues
! Free cooling year round ! Use modular to right-size and scale quickly ! Green
– CSC buys certificates of carbon neutral energy – 100% Finnish hydro power
Outside
Timeline
Paper mill
2010 2011 2012
Site selecGon
Planning
Warehouse
2013
1st ITT
Datacenter
Supercomputer
October
2nd ITT
Analysis
Delivery Conven&onal
build
Modular build
WHAT DID WE BUILD?
CSC’s new Kajaani datacenter
! Renovated paper warehouse ! 12 000 m2 of space ! ~ 1.7 x Old Trafford football ground
Warehouse has two DC tenants
Hub datacenter
SGI Ice Cube R80
TWO DATACENTERS IN ONE
3D WALKTHROUGH VIDEO
Specification
! 2.4 MW combined hybrid capacity ! 1.4 MW modular free air cooled datacenter
– Upgradable in 700kW factory built modules – Order to acceptance in 5 months – 35kW per extra tall racks – 12kW common in
industry – PUE forecast < 1.08 (pPUEL2,YC)
! 1MW HPC datacenter – Optimised for Cray super & T-Platforms
prototype – 90% Water cooling
Hub building and DC
Hub datacenter
Hub building
! Redundant network rooms ! 100kW UPS – upgradable to MW’s ! 10KV switchgear ! Fire suppression ! Storage rooms – warehouse is cold ! Hub Datacenter
Hub Datacenter
Chillers 2 x 450 kW
External dry coolers
Hub DC facts
! Due November ! 900kW water cooling
– + 100kW air from hub
! Purpose built for water cooled HPC
IT summary
• Cray “cascade” supercomputer – 10M€ -‐ five year contract – Fastest computer in Finland due mid-‐November – Phase one 385 kW 2012 of Intel processors – Very high density, large racks
• T-‐Plamorms prototype – Very high density hot-‐water cooled racks – Intel processors, Intel and NVIDIA accelerators – TheoreGcal 400 TFlops performance
SGI Ice Cube R80
SGI Ice Cube R80
SGI Ice Cube R80
! One head unit and two expansion modules ! More modules can be added ! Fully automated free cooling system
– Dozens of cooling fans, louvers and sensors ! Extremely energy efficient – pPUE 1.08 ! Set point allowed to vary (10-27C for us) ! Adiabatic cooling on warm days ! Exhaust heat used to warm incoming air
Factory visit August 2012
! “Super Cluster” – 4.5M€ five year contract – 1,152 Intel CPUs – 190 TFlop/s – 30kW 47U racks
! HPC storage – 3PB of fast parallel storage – Supports Cray and HP systems
IT summary
CONSTRUCTION
Site in January 2012
MDC TIME LAPSE VIDEO
Head load test
• 700kW of load banks – 20 racks – 120 x 6kW load banks – 2 days to rack mount – Half MDC capacity
• pPUE 1.05 • Very useful • Silenced the criGcs
Commissioning
! For MDC it took longer than construction!
! MDC internal workings and drawings are sensitive – Test plan process challenging – Training required for new free
cooled system type ! Where possible plan in
advance ! Allow plenty of time
LESSONS LEARNED
Lessons learned
! Simplify your procurement! ! Share every detail of your site with bidders ! A modular datacenter is still a datacenter
– Even if design fixed is expect many decisions ! Pay special attention to integration
– fire, security, power, electricity, etc. – Any small issue can cause major delay
! Get expert advice – don't assume!
Reference or define every ’standard’ Don't assume
! Racks – Define depth, height, RU, posts – Strongly consider alternatives like
opencompute ! Door heights ! Make a list of required standards
– Fire suppression – insurance, building permit ! Get a datacenter contract
– Modular buildings contracts != IT contracts – If possible multiple phases of acceptance
Things we got right
! Make the selection based on 10 year TCO – IT TCO was 5 years per phase – highlights energy efficiency without defining 'green’ – could also use an artificially high energy price
! Require that IT vendors list – PDU efficiency use 80Plus – Temperature and humidity ranges – ASHRAE TC 9.9 2011
! All our suppliers support A2 range (5 to 35C)
Datacenter operations team
QUESTIONS? Thanks
More information
! Contact me – peter.jenkins@csc.fi
! Blog: – http://www.csc.fi/blogs/gridthings