Your SlideShare is downloading. ×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×

Saving this for later?

Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime - even offline.

Text the download link to your phone

Standard text messaging rates apply

Visualization in the Design and Operation of Efficient Data Centers

1,256
views

Published on

Published in: Technology, Business

0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
1,256
On Slideshare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
12
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. VISIT 2010 – Fujitsu Forum Europe 1
  • 2. Innovation Forum – Room 12 Shaping Visualization in the Design and Operation tomorrow with you. of Efficient Data Centers Dr. David Snelling Research Project Manager Computer Architectures, Fujitsu Laboratories 17:15 hVISIT 2010 – Fujitsu Forum Europe
  • 3. 3
  • 4. Visualization in the Design andOperation of Efficient DataCentersDr. David SnellingFujitsu Laboratories f EuropeF jit L b t i of E Copyright 2010 Fujitsu Laboratories
  • 5. Background g The reduction of energy consumption and CO2 emissions are serious environmental issues. issues Energy spent on non-IT functions, e.g. air conditioning and power management, need to be reduced. IT equipment efficiency needs improving. 250 sumptionh) 消費電力量(億kWh)Energy rgy Cons (x108kWh 140 Floor space Space 総床面積(万m2) or 4㎡) 120 200 e(x10 100 150 on Air conditioner IT equipment Ene consumptio 80 0 Floo 60 100 IT resources IT facilities Power unit 40 50 F 20 Lighting 0 0 2007 2008 2009 2010 2011 2012 2007 2008 2009 2010 2011 2012 (FY) Ratio of energy consumption in IDC Energy consumption of IDC in Japan (Fujitsu, FY2006) (MIC Research Institute 2008) VISIT 2010 Fujitsu European Forum 5 Copyright 2010 Fujitsu Laboratories
  • 6. Holistic Approach pp Data Centre 100% 100% Data Centre input CERES Equipment 40% 25% 35% 40% IT input Power Cooling Servers 65% 15% 20% Network 26% Servers Storage CPU 30% 25% 45% 7.8% CPU Power Others Utilisation 20% 80% 1.6% 1 6% Work IdleVISIT 2010 Fujitsu European Forum 6 Copyright 2010 Fujitsu Laboratories
  • 7. CERES Motivation Cost and Energy gy Reduction Evaluation 3% 9% Power Service 59% 29% Cooling IT Other Aims: Reduce energy use in data centers 3% 9% Quantify savings in advance Power of decision making 29% Cooling IT 59% Predict effects of operational Other changes to facilitate risk analysis Identify precisely where energy losses existVISIT 2010 Fujitsu European Forum 7 Copyright 2010 Fujitsu Laboratories
  • 8. Holistic Simulation WHY Data center energy and heat schematic RESULTS Soaring energy energ Visible ¥ and CO2 costs impact promotes best practice CO2 is a regulatory DC lifetime cost & issue environmental impact known for CSR all services Energy New green constraints services Final results of this example show reduction in cost and CO2 by using more efficient chillers HOW HOW HOW Mathematically link IT use to Build on existing technologies to Derive immediate customer direct & indirect energy use predict actual costs value from reduced energy useVISIT 2010 Fujitsu European Forum 8 Copyright 2010 Fujitsu Laboratories
  • 9. Example Outcomes Completed a CO2 Value Analysis for a UK DC Simulator validated: 95% to 98% accurate Action Recommended for UK DC by % CO2 Annual Cost CO2 Value Analysis Reduction Savings Shutdown air conditioning and UPS g 1.4% 1 4% £8K / Y Year when disaster recovery hall not in use Use exactly the right air conditioning in 9% £52K / Year main computer halls Add free cooling to data center 17% £98K / Year Use variable speed fans in computer 14% £81K / Year room air conditioners All the above 29% £167K / Y YearVISIT 2010 Fujitsu European Forum 9 Copyright 2010 Fujitsu Laboratories
  • 10. Server Level Data Centre 100% 100% Data Centre input Equipment 40% 25% 35% 40% IT input Power Cooling Fiber Optic Servers 65% 15% 20% Network 26% Servers Storage CPU 30% 25% 45% 7.8% CPU Power Others Utilisation 20% 80% 1.6% 1 6% Work IdleVISIT 2010 Fujitsu European Forum 10 Copyright 2010 Fujitsu Laboratories
  • 11. Real-Time Multiple-Point Measurement Temperature distribution is easily and p p y precisely y measured by observing Raman scattering light 10,000 10 000 data points along 10,000m single 10 000m optical fiber Principle of measurement Data Centre Temp. Temp Laser pulse Optical Fiber Laser Raman Scattering Light Detector Servers Optical fibre Intensity of Raman scattering light Propagation time Floor Temp. of points in optical fiber DistanceVISIT 2010 Fujitsu European Forum 11 Copyright 2010 Fujitsu Laboratories
  • 12. Temperature Distribution p 2D temperature distribution of server rack surfaces Hot spot 24 21 Temperature distribution along an optical fibre erature Tempe Location Knowing the precise temperature distribution across racks every 20 seconds facilitates emergency response An energ sa ing of 10% to 30% b optimi ing air conditioning energy saving by optimizing air-conditioningVISIT 2010 Fujitsu European Forum 12 Copyright 2010 Fujitsu Laboratories
  • 13. Device Level Data Centre 100% 100% Data Centre input Equipment 40% 25% 35% 40% IT input Power Cooling Servers 65% 15% 20% Network 26% Servers Storage CPU 30% 25% 45% 7.8% CPU Power Others GaN HEMT Utilisation 20% 80% 1.6% 1 6% Work IdleVISIT 2010 Fujitsu European Forum 13 Copyright 2010 Fujitsu Laboratories
  • 14. GaN HEMT Technology gy New transistor in Fujitsu Labs. The worlds highest on-state current density while achieving a turn-on voltage of +3 V, making it the first GaN-HEMT in the world that has the characteristics required for power supply. Target area for power supply Why GaN for power supply applications ? ty 1200 urrent densit Electrode Electrode A/mm) Si 800 (mA On-state cu GaN Smaller transistor sizes became possible. 400 other reports ・ Breakdown voltage: 10 times higher Small size ・ Electron velocity: 2 times higher 0 Low loss 0 2 4 6 ・ Electron density: 10 times higher Turn-on voltage (V)VISIT 2010 Fujitsu European Forum 14 Copyright 2010 Fujitsu Laboratories
  • 15. Effect of GaN HEMT Power loss can be reduced by one-third one third compared to power supplies based on conventional silicon transistors transistors. Silicon Transistor GaN Transistor Switching -loss Other losses On Power - Other losses saving -loss l (i.e., Transformer)VISIT 2010 Fujitsu European Forum 15 Copyright 2010 Fujitsu Laboratories
  • 16. Gallium Nitride (GaN) HEMT ( ) Power-saving data center Data-center power consumption in Japan could be reduced by 12% (330,000 tons CO2) Small-size power supply Smaller power supplies would contribute to reducing space requirements in servers The size of AC adapters for notebook PCs could be reduced to one tenth current sizes – “Brickless” laptops one-tenth Brickless GaN Transistor Power Supply Applications Server Data Centre LaptopVISIT 2010 Fujitsu European Forum 16 Copyright 2010 Fujitsu Laboratories
  • 17. Where to Next? Power-saving server Power-saving laptop PC with small AC adaptor Power plants Power-saving Power saving home electronics Industrial Lifestyle innovation machines Motors mobile phone bil h Electric vehicleVISIT 2010 Fujitsu European Forum 17 Copyright 2010 Fujitsu Laboratories
  • 18. Service Level Data Centre 100% 100% Data Centre input Equipment 40% 25% 35% 40% IT input Power Cooling Servers 65% 15% 20% Network 26% Servers Storage CPU 30% 25% 45% 7.8% CPU Power Others Utilisation 20% 80% 1.6% 1 6% Work Idle SysVizVISIT 2010 Fujitsu European Forum 18 Copyright 2010 Fujitsu Laboratories
  • 19. SysViz: Motivation y System performance has suddenly degraded, but how to locate the cause? It will take hours to cross- check the logs of components (network, database, middleware)! Open, multi-vendor, huge and complex systems We have to solve it quickly in order to meet SLA! Challenge: View and understand the interactive behavior of large numbers of heterogeneous components in real-time • See where current and future problems/opportunities exist • Manage them appropriately • Retain agility to meet customer demands at the service levelVISIT 2010 Fujitsu European Forum 19 Copyright 2010 Fujitsu Laboratories
  • 20. SysViz: Deployment y p y Non-intrusive, agent-less monitoring , g g Network traffic directly captured from the IP network by p y port-mirroring network devices g • no need of agents on servers nor software modification • no risk of system performance/reliability degradation Just plug SysViz engine It can be applied to operational systems to the network device without affecting its operationVISIT 2010 Fujitsu European Forum 20 Copyright 2010 Fujitsu Laboratories
  • 21. SysViz: Configuration y g Automatic system behavior modeling System behavior model gives the sequence of messages that compose each business transaction • generated by our event mining technology that identifies • correlations and causalities between network messages Crude Target System S stem Behavior Model B h i M d l Response breakdown 450 esponse Time [ms] 均サーバ処理時間 [ms] 400 350 300 250 Auto Generate Report 200 150 100 平均 Re 50 0 9:30 9:31 9:32 9:33 9:34 9:35 9:36 9:37 9:38 9:39 9:40 9:41 9:42 9:43 9:44 9:45 9:46 9:47 9:48 9:49 9:50 9:51 9:52 9:53 9:54 9:55 9:56 9:57 9:58 9:59 Frequency 時刻 Web/AP DB BTOnline(HTTP) BTOnline(DB) erroneous causalitiesVISIT 2010 Fujitsu European Forum 21 Copyright 2010 Fujitsu Laboratories
  • 22. SysViz: Operation y p Real-time model based monitoring Refine a crude model manually during operations to represent y g p p normal behavior of business transactions. Network traffic is filtered by using this behavioral model • Unusual be a o that de a es from the model U usua behavior a deviates o e ode • raises alerts in the monitoring dashboard Crude Refined Target System Behavior Model B h i M d l Operation Manager This component is slower than Auto Generate Visualize usual These components show unusual operation flow erroneous causalitiesVISIT 2010 Fujitsu European Forum 22 Copyright 2010 Fujitsu Laboratories
  • 23. Holistic Summary y Data Centre 100% 100% Data Centre input CERES Equipment 40% 25% 35% 40% IT input Power Cooling Fiber Optic Servers 65% 15% 20% Network 26% Servers Storage CPU 30% 25% 45% 7.8% CPU Power Others GaN HEMT Utilisation 20% 80% 1.6% 1 6% Work Idle SysVizVISIT 2010 Fujitsu European Forum 23 Copyright 2010 Fujitsu Laboratories
  • 24. VISIT 2010 Fujitsu European Forum 24
  • 25. VISIT 2010 Fujitsu European Forum