Passively Cooled Data Centers (Weec 2009)

Passively Cooled Data Centers

Jamil Scott

Data CenterData

Center

Key Cause & Effect RelationshipsData Center

Data Processing

Energy Consumption

Profits

Key Cause & Effect RelationshipsData Center (Beneficial Effects Only)

Electricity

Processors

Data

Data Processing

Power Supply

Processor Power

Ease of Manufacturin

g

Profits

Lower Cost Servers

Key Cause & Effect RelationshipsData Center (With Harmful Effects)

Electricity

Processors

Data

Data Processing

Excessive Heat In

Processors

Power Supply

Processor Power Excessive

Heat Near Processors

Processor Damage

Ease of Manufacturin

g

Profits

Lower Cost Servers

Close Proximity

Key Cause & Effect RelationshipsData Center (With Harmful Effects & Counteractions)

Electricity

Processors

Data

Data Processing

Excessive Heat In

Processors

Processor Damage

Power Supply

Processor Power Excessive

Heat Near Processors

HVAC

Cooling

Low Density Cooling

Electricity Consumption For Cooling

Ease of Manufacturin

g

Profits

Lower Cost Servers

Cooled Air

Close Proximity

CoolingCooled

Air

Data Processing

Ease of Manufacturin

g

Power Supply

Key Cause & Effect RelationshipsData Center (Full System View)

Electricity

Processors

Data

Excessive Heat In

Processors

Processor Damage

Processor Power Excessive

Heat Near Processors

HVAC

Electricity Consumption For Cooling

Low Density Cooling

Profits

Lower Cost Servers

Close Proximity

Data Processing

Power Supply

CoolingCooled

Air

Ease of Manufacturin

g

Cooling Mechanism Cooling is provided to counteract excessive heat, but

requires consumption of electricity Current cooling methods rely on cool air for cooling. Air is

significantly less dense than processor materials. The disparity between materials makes air cooling relatively inefficient.

Form Factor Power supplies are placed in close proximity to processors

to provide ease of manufacturing & installation Power supplies generate (and can operate in conditions

of) heat in excess of processor limits Processor Construction

Current processor construction allows the processor to generate sufficient heat to damage itself.

Key System Contradictions

Page 7

Find an alternative way to obtain: Cooling that offers the following: Eliminates, reduces, or prevents High Heat Near Processors and Excess Heat in

Processors Does not cause Additional Electricity Consumption Does not require HVAC and Air Is not influenced by Low Density Cooling.

Proposed Resolutions Eliminate HVAC

Use a dense cooling material (water, oil, etc..) Use a dense thermal transfer material (metal, concrete, etc…) Use conduction to move heat away from processor boards

Then forced convection with flows other than air Find a beneficial use for the excess heat

Power generation power via sterling engine

Key System Contradictions(Cooling Mechanism)

Page 8

Find an alternative way to provide Processor Power without causing High Heat Near Processors.

Proposed Resolution Separate the power supply from the processor boards Provide power through a standardized power bus Provide data through a standardized data bus

Find a way to eliminate Rack-based Server. Proposed Resolution

Redesign processor boards for: Flush mounting against processor surface Machine (e.g. robot) handling

Standardize power and data connections for machine handling Make the data center the ‘rack’

Use the thermal transfer structure to provide the mounting surface Minimize space between walls to provide dense processing capacity

Key System Contradictions(Form Factor)

Page 9

Convection Concept

Page 10

What changes will be required? Redesign Processor boards Standardize power connections Standardize data connections Design machine-based board management system Re-engineer data centers

Redesign spaces to provide optimum environment for processing components, not humans

Data Center becomes the server Walls become the rack and heat transfer mechanism Reuse waste heat

Making Passive Cooling a Reality

Page 11

Current Processor Board Design

Page 12

Ease of Manufacturin

g

Power Supply

Key Cause & Effect Relationships (Revisiting the Full System View)

Data Processing

Ease of Manufacturin

g

Power Supplyb

ElectricityClose

Proximity

Processors

Data

Excessive Heat In

Processors

Processor Damage

Processor Power Excessive

Heat Near Processors

HVAC

Low Density Cooling

Electricity Consumption For Cooling

Profits

Lower Cost Servers

Close Proximity

Data Processing

Power Supply

CoolingCooled

Air

Ease of Manufacturin

g

Passive Coolin

g

Key Cause & Effect Relationships(Improved Ideality)

Electricity

Processors

Data

Data Processing

Excessive Heat In

Processors

Processor Damage

Power Supply

Processor Power

Passive Coolin

g

Profits

Lower Cost Servers

Close Proximity

Ease of Manufacturin

g

Separating the power supply from the processors allows each to operate in their own optimized environment

Flush mounting the processor boards allows natural convection to remove excess heat without electricity consumption

Redesigning processor boards for flush mounting & machine handling allows re-thinking board design High speed inter-board communications (optics, etc…)

Redesigning processor boards allows re-thinking of data center architecture Data Center is the server Significant increase in processor density in proportion to floor space Dynamic reallocation of processing capacity Enhanced security by optimizing the environment for systems rather

than people Very tight tolerances between boards Elimination of air (to reduce thermal radiation) Reuse thermal energy

Benefits

Page 15

Rethinking data center, processor, and server designs will open up a new and exciting period in information processing.

Meeting the processing needs of the future requires new thinking Future data centers must be designed to host

processors, not people. Future processors must use less energy and

produce less heat Processing components must operate in optimized

environments

Conclusion

Page 16

Jamil ScottPrincipal Information Systems Engineerjamil@mitre.org(703) 983-5709

Dr. Thomas J. HeimanSenior Information Systems Engineertheiman@mitre.org(703) 983-2951

Team Members

Page 17

Questions

Page 18