Thinking inside the box: Boosting the effectiveness of air coolingStory
May 09, 2008
Cooling is an integral part of any system, and particularly those used in harsh environments. David discusses cooling system parameters as applied to an industrial computer installed in a military environment where the high ambient temperature and airborne dust will impact system reliability.
Cooling is an integral part of any system, and particularly those used in harsh environments. David discusses cooling system parameters as applied to an industrial computer installed in a military environment where the high ambient temperature and airborne dust will impact system reliability. Equations describing electronic component life versus temperature and system airflow, along with inductance, are examined. A case study is also presented, illustrating that internal system temperature considerations are critical.
System software performance continues to be a critical factor in the embedded military space. In the end, however, environmental issues will often determine the success or failure of integrated computing platforms installed in mission-critical applications such as Command and Control, modeling and simulation, and communications. Computers intended for military applications are subjected to a broad range of punishment, including elevated ambient temperature. While shock, vibration, and power quality typically cause immediate failure, operating at elevated temperatures is more insidious, shortening the system's Mean Time Between Failure (MTBF). Studies have demonstrated that component life is typically reduced by 40 to 50 percent for every 10 ¬∞C increase in operating temperature. Thus, it is paramount to reduce the temperature of all components inside a computer system as much as possible. Filtering out the environment, calculating the airflow, and factoring in impedance are vital to conquering this challenge.
The high-temperature challenge
In understanding the effects of increased temperatures, system life expectancy decreases exponentially with increased temperature. The extent of this degradation is generally governed by the Arrhenius equation (below), which describes how component age accelerates with increasing temperature.
Table 1: Arrhenius Equation
(Click graphic to zoom by 1.6x)
Simplified, this equation supports the rule of thumb that component life halves for every 10 ¬∞C rise in component temperature. Thus it is easy to see that MTBF goes down significantly with any increase in temperature beyond the original calculations.
Providing sufficient cooling, especially for high-power multiprocessor systems in the present combat theater, requires a disciplined approach to cooling system design. The end result is a combination of several smaller contributions where any degradation in any one will impact system reliability.
Filtering out the environment
Military grade computers expected to operate in ambient temperatures in excess of 120 ¬∞F will employ multiple fans mounted either at the front or near the center of the chassis as dictated by the installed components. These configurations provide high airflow across a wide area as well as a positive pressure within the chassis. Mounting fans at the rear of a chassis as exhaust fans induces a negative pressure inside the chassis, which will pull unfiltered dirty air in through any opening in the chassis.
A filter should be installed in the inlet path to remove dirt from the air stream. Even small amounts of dirt, in combination with moisture, can be conductive and cause intermittent operation. Dirt acts as a heat insulator and, even in small amounts, will somewhat degrade cooling effectiveness. In larger amounts, dirt can completely obscure the passages in the heat sinks on the processors and chipsets, reducing the cooling to a small fraction of that required by these devices. Any insulative effect will raise component temperatures and lead to a shorter life. Software operation may also be impacted if the processor goes into throttle mode because of overheating, automatically reducing the clock rate to reduce power consumption and heat generation.
Specification of the filter media and size is important because a filter reduces the airflow through a system. A trade-off in filter efficiency versus airflow is required to assure adequate system cooling flow. Fans with higher-pressure capability should be specified for use with filters. Filters also need regular cleaning as the trapped dirt will increase the pressure drop across the filter. A filter can become completely blocked, reducing airflow through the chassis to almost zero.
An important design criterion is completely sealing the fan bulkhead including cable passages to prevent air recirculation within the chassis. Improper sealing will allow higher-pressure hot air in the rear of the chassis to circulate around to the inlet side of the fans, significantly degrading the cooling effectiveness. Not only is the airflow through the chassis reduced, the air being circulated through the chassis will be hotter than external air. These openings do not have to be very large, on the order of a couple of square inches combined area, to reduce the airflow through the chassis to almost zero. The result is a chassis poorly cooled only through the skin and the small amount of flow through the power supply.
Figuring the airflow
One factor to consider is the impact of using smaller fans, such as in cases where chassis height must be reduced. Smaller fans provide reduced airflow in a nonlinear fashion. Examining the Fan Laws, which describe the basic relationship between fan speed, flow, pressure, and power, shows Q ~ ND3, where Q is airflow, N is speed, and D is diameter. Thus, airflow is governed by the blade diameter cubed. That is, a fan half the size will move one-eighth the air for the same RPM. Alternatively, half-size fans must theoretically spin eight times faster to deliver the same amount of air. However, there is a limit to fan speed with the result that fans with adequate flow rates may not be available for high-power systems. Higher-speed fans are louder, and the higher frequency makes them seem noisier. This can be a critical factor when multiple systems are operating in a confined space and there is a strict noise limit imposed. Therefore, systems should be designed with the largest fans possible given space constraints.
Table 2: Calculating the airflow requirements in a configured system
(Click graphic to zoom by 1.6x)
Note that ambient temperature does not enter into the equation. Higher ambient temperatures will somewhat reduce the output of the fans due to the lower density, but this is a negligible effect. Ambient temperature is important as it provides the baseline for the system temperature. If the ambient temperature rises 20 degrees above the tested temperature, the internal temperature will rise the same amount.
A fan requirement determination first requires an analysis to find the system component with the lowest maximum temperature rating. Generally this is a disk drive, but these are typically installed directly in the incoming air stream and not heated by other system components. But it might also be a plug-in card that will be mounted near and downstream of the hot processors. Or it could be a processor with a small margin to maximum operating temperature. Thus, if a component has a 125 ¬∞F limit and the system will be used in an ambient temperature of 115 ¬∞F, then the allowable temperature rise through the system, conservatively, would be 10 ¬∞F.
Table 3: Airflow through a system
(Click graphic to zoom by 1.6x)
This is the airflow through a system, not the free rating of the fan. Even in a well-designed system, fans will output at most 50 to 60 percent of their rated flow. Thus, in this case, a starting point of 300 CFM would be indicated for initial fan selection.
An important point to note is placement of the power supply, in regard to airflow. A rear-mounted supply will exhaust its hot air directly out the rear of the chassis. A front-mounted power supply will exhaust into the chassis. In general, a switching power supply is approximately 70 to 80 percent efficient. This inefficiency needs to be added to the total watts dissipation calculations. System components using 500 W, for example, would lead to total system power usage of 500/.70 = 714 W for a 70 percent efficient supply. This is the number to use for a CFM calculation, but only for a front-mounted supply.
Chassis flow impedance
Another challenge when selecting fans is to understand chassis flow impedance, which will determine the volume of air that specific fans can push through a particular enclosure. Chassis flow impedance is defined as the resistance of airflow from the point of intake, through the enclosure, to the point of exhaust. Size and number of vent openings, filter media, internal structure, and installed components all contribute to increased impedance. Due to the infinite variations presented in chassis design and the configuration of installed components, the impedance can only be discerned by measuring the actual flow versus the required pressure. Mathematical tools such as Computational Fluid Dynamics (CFD) can be used to approximate flow. However, a measurement of the actual chassis impedance should be performed to validate system performance and the CFD model.
Addionally, pressure gradients drive airflow. There is low pressure in front of the fans with higher pressure outside the chassis so air flows into the chassis. The fans boost the pressure above ambient so air then flows into the rear and out the back of the chassis. Physics shows that airflow is proportional to the square of the pressure. That is, to double the flow through a system requires four times the pressure. All things being equal, adding a second fan will not double the flow but will increase it, at most, by the square root of 2 (1.41) or about 40 percent. To double the airflow would require four times the pressure.
Flow through a chassis can be approximately determined by laying a fan curve over a chassis impedance curve. The intersection of the two curves gives the airflow.
Figure 1 plots impedance curves for several different chassis. A restrictive configuration with high impedance is shown as 'A' while an open, well-ventilated chassis is shown as D. B and C show impedance curves for chassis with intermediate impedance ranges. A single fan (black line), two series fans (red line), and two parallel fans (blue line) are overlaid on the impedance curves. The amount of air that will flow through the 'D' chassis can be seen for the three fan configurations, ranging from 66 CFM for a single fan to only 71 CFM for two fans in series. A dramatic improvement to 92 CFM is shown for two fans in parallel.
(Click graphic to zoom by 2.0x)
It is evident from this graph that for a chassis with high impedance, two fans in series provides better cooling; meanwhile, for an open chassis with low impedance, two fans in parallel flow better. The B and C curves show very little difference between series and parallel fan installation, though there is improvement over a single fan.
Note that D impendence curve intersects the parallel fan curve (blue line) at the area of fan instability where minute changes in pressure have a large effect on airflow. Selection of different fans would be indicated in this situation to assure more determinate airflow.
As the airflow increases with an additional fan, the internal pressure also increases, thus reducing each fan's output. Note the fan curve for the second fan in series simply duplicates the curve for a single fan with double the flow for each pressure point. There is no additional pressure available. The same applies to fans in parallel, except flow doubles with no increase in available pressure.
To visualize the effect this would have on chassis flow, imagine a chassis with zero impedance, essentially an open box. Since there is no back pressure, two fans in series won't move more air than a single fan, but two fans in parallel would double the flow. At the other extreme, imagine a chassis with very high impedance, such as a closed box with just a small opening. Two fans in parallel will not increase the pressure, so no additional air would flow. On the other hand, two fans in series would double the pressure, giving 40 percent more flow through the system. In the real world, chassis impedance is somewhere between these extremes, so that fan selection and configuration is determined by comparing the curves to determine best flow. Another consideration is that fans are not linear devices and have areas of instability. Careful selection of the fans based on the intersection of the curves will prevent operation in these areas of instability and maximize flow through the system.
Case study: 4U Joint Range Extension system
As a real-world example of cooling analysis and solution for a high-power military computer, consider the problem of installing two single board computers inside one 4U enclosure. When L-3 Com ESD needed a rugged 4U enclosure to support its Joint Range Extension (JRE) program, Chassis Plans engineers designed a comprehensive solution tailored to JRE's unique mechanical and environmental system requirements. Joint Range Extension is a combined hardware and software system that receives battlefield information transmitted on a tactical data link in a particular area of operations, then forwards that information to another tactical Data Link Terminal (DLT) located at a point beyond the line of sight.
Chassis Plans' JRE-DLT (Figure 2) provides two dual processor Xeon XPT single board computers, one running Windows XP and the other Solaris. Due to the level of heat generated by hosting two computers in one enclosure, the remedy for the JRE-DLT enclosure was to create a "wall of air" by mounting four high-velocity 92 mm hot-swap fans mid-chassis and sealing the airflow path to eliminate recirculation. Because this system is used in very dirty environments, a pair of 30 ppi reticulated polyurethane foam filters were provided on the front door.
(Click graphic to zoom by 1.2x)
The system includes two independent backplanes that occupy the full width of the chassis. This dictated the use of a front-mounted power supply. System power requirements were calculated at 450 W. With a front-mounted power supply with 78 percent efficiency, system heat generation was calculated to be 576 W. With a target temperature rise of 15 ¬∞F, target system airflow was calculated to be 121 CFM. The fans were rated for 76 CFM each, for a total airflow of 305 CFM free air rating. System impedance reduced the fan output to approximately 150 CFM, meeting the target system cooling parameters with flow margin. Careful attention to chassis design details, in particular airflow paths and fan selection, provided sufficient cooling to assure adequate performance and component life from this high-power system.
Sidebar 1: Another low-temp alternative
(Click graphic to zoom by 2.2x)
Cooling electronics down when temperatures rise
Punishing thermal environments and demand for reliable system operation require a careful analysis of airflow through a military computer system. If component life is halved for every 10 ¬∞C rise in operating temperature, then it is imperative to minimize component temperatures by maximizing cooling airflow. Optimum fan selection depends on the flow resistance imposed by chassis design, air filter media, and installed components; additionally, the best fans to use can be determined by plotting chassis impedance versus the fan flow/pressure curves. Only through complete system engineering including cooling can a system's operational potential be realized.
David Lippincott founded Chassis Plans in 1997 as a design firm specializing in rugged industrial computers. In 2001, he reformed Chassis Plans with his partner, Steve Travis, in order to manufacture and integrate those same types of computers, with special emphasis on airflow, heat dissipation, and shock and vibration mitigation. Prior to founding Chassis Plans, David was cofounder and engineering vice president of Industrial Computer Source, a company that manufactured industrial computers and products. He has a degree in Aerospace Engineering Science from the University of California, San Diego, and has spent his career in industrial computing systems, industrial control, and computer design. He can be reached at [email protected].