John Daily Bill Best Nate Slinin
Project Concept Our focus is centered around addressing the growing demands placed on the cooling infrastructure in a modern data center.
Facing the Truth Computational efficiencies over the last 40 years have doubled every 1.6 years providing a smooth logarithmic growth which is derived from Moore's Law However power use doubled between 2000 and 2005 when demand for computational power was a fraction of today’s requirements.
Current Method Centered around the Computer Room Air Handler. Image:
Traditional Method Using CRAH system Immersive Liquid Cooling System ILC Exchanger
Benefits over Forced Convection Liquid has higher thermal conductivity than air. Better electrical insulator. Its high heat capacity decreases risk of hardware failure in case of cooling failures
Key ILC Design Challenges Non-Conductive Fluid
The effectiveness of heat transfer and heat dissipation of the fluid will be determined by its thermal conductivity and the cooling requirements of the most critical portion of each server, the CPU
To measure the amount of heat production we can transmit through the fluid we set the thermal source as the chip junction temperature, and each exchange point is modeled as resistance leading to the ambient liquid temperature. Since thermal resistance is additive we can add the various junctions together to get the total resistance and determine the ambient temperature of the fluid in its passive state. This is given by the equation R= ∆T/ P th – R source ; where R is the maximum thermal resistance, ∆T is the change in temperature (T j -T L ) and P th is the thermal power in watts.
Key ILC Design Challenges Non-Conductive Fluid Cooling Configuration
Replacing traditional data center 19inch CEA- 310-E racks with a enclosure that will house standard 19 servers in 1U, 2U or 4U configurations. Heat exchange plates will be placed in between each server.
Our rack equivalent enclosures will be placed horizontally as opposed to standing vertically and server will be inserted from the top at waist level.
Key ILC Design Challenges Non-Conductive Fluid Cooling Configuration Control System
PID CONTROLLERSERVER - + T_DT_A error Control System Model
MICROCONTROLLERWATER PUMP PID CONTROLLER
PID Equation PID = K_P*e+K_I*∫edt+K_D*(de/dt) Constants K_p = 0.15 K_i = 0.05 K_d = 0.05
timeT_DT_OeK_PK_P*eint(e)K_IK_I*int(e)deriv(e)K_DK_D*deriv(e)PIDT_SCooling Ave31.87Total Max40.00Ave3.38 Max5.86
The Future Concurrent Systems Image:
Concurrent Systems Chip manufactures are pushing new CPU packages with multiple processors. GPU's are becoming increasingly like CPU's and may have helped fuel the popularity of distributed computing, Map Reduce, Software as a service models and other concurrent systems in the marketplace.
The Future Concurrent Systems Sustainability Image:
Sustainability By reducing the number of exchanges the increase in heat transfer can be passed to an external environment can include active sources currently used today such as a water chiller or passive sources such as a large body of water or geothermal heat pump.
The Future Concurrent Systems Sustainability Flexibility and Challenges Image:
Flexibility and Challenges Greater utilization of current datacenter space and the ability to retrofit our system into current outdated water-chilled datacenters. Current industry stigma against change. The IT industry doesn’t like radically different approaches which are viewed as risky. The project only becomes feasible when ROI is high enough to offset risk.