Here is the abstract you requested from the Thermal_2014 technical program page. This is the original abstract submitted by the author. Any changes to the technical content of the final manuscript published by IMAPS or the presentation that is given during the event is done by the author, not IMAPS.
|Evaluating liquid cooling at the rack: Comparing distributed and centralized pumping|
|Keywords: hybrid warm liquid cooling, energy efficiency, rack-level system|
|Continuing growth of the data center industry has exerted a visible strain on the power grid over the last decade. Recent surveys report that a majority of data centers in North America operate extremely inefficiently with IT equipment accounting for less than half the total power consumption (support infrastructure responsible for majority of the remaining). With cooling systems contributing a significant portion of this overhead, it has become vital that energy savings and efficiency be pursued in these components at various levels within the data center facility. In addition, increasing power densities at module and server levels are pushing the limits of air cooling for energy-efficient and reliable operation. In such situations liquid cooling serves as a better alternative with higher heat capacity, directed cooling and lower pumping power. In this study, 2OU (OpenU) web servers are tested across the ASHRAE liquid cooling envelope and the effect of higher inlet temperatures in terms of IT and cooling powers, and internal component temperatures are reported. Nine web servers are retrofitted with a solution that provides indirect liquid cooling of CPUs using active cold plates and recirculated air cooling of remaining components with a fan-assisted radiator. In essence, each server is completely liquid cooled as the coolant entering the chassis passes through the radiator and two cold plates in series before exiting. These servers were then installed in a short, modified Open Rack in a test-bed data center facility. The rack is equipped with two heat exchangers that, in series, exhaust heat from the servers to the environment. Considering one of the wider environmental envelopes proposed by ASHRAE for liquid cooling, W4, the servers were subjected to inlet temperatures varying from 20°C to 45°C. In order to simulate a variation of representative workloads, at a fixed inlet temperature, each server was equipped with a script that (in tandem) stressed the system at power levels varying from idling to 100%. Failure scenarios, common in a working environment, are simulated by unhooking a pump and a fan (both in isolation and together) from the server operating at the highest temperatures (determine from previous tests). Thus, for a given pumping system, comprehensive testing is carried out for both regular and impaired operation and performance parameters such as device temperatures and power consumptions are reported. The overall system is then modified to evaluate centralized pumping by disabling pumps in all cold plates and installing larger redundant pumps at the rack. Aforementioned tests are conducted from this setup and suitable comparisons between the two pumping systems are reported.|
|John Fernandes, PhD Candidate
University of Texas at Arlington