On demand cooling with real time thermal information

In recent years, the internet services industry has been developing rapidly. Accordingly, the demands for compute and storage capacity continue to increase and internet data centers (IDCs) are consuming more power than ever before to provide this capacity. With large scale power usage comes the large price tag of paying for that power. With growth trends increasing and development expanding, IDC owners realize that small improvements in efficiency, from architecture design to daily operations, will yield large cost reduction benefits over time. One important trend for reducing power consumption is raising the operational temperature of an IDC. Higher ambient temperatures (HTA) have been shown to reduce power consumption based on more efficiently using computer room Air conditioning units. This paper outlines the key architectural details around a customized server and data center design called on-demand cooling. It describes how IDC owners can utilize server real-time thermal data, including inlet/outlet temperatures, power, and server fan speed (airflow demand) as key inputs for improved IDC cooling control and HTA operation. First off, we need to understand current limitations of traditional IDC's and HTA. In most traditional IDC's, the operational temperature range continues to be between 18-23C or even lower. Here are three reasons to explain why conservative cooling settings continue to be used. The first reason for conservative cooling settings (over cooling) was needed in the past, was to ensure all IT infrastructure was cooled to within equipment specifications. Because these specifications varied widely between different IT infrastructures, there was the potential to impact the reliability of certain IT equipment with high temperatures. Successful high temperature ambient (HTA) data center operations is accomplished by understanding all equipment temperature thresholds and keeping the ambient temperature within those reliability levels, to still meet SLA and business application requirements. Fortunately today, modern IT infrastructure (servers and switches) are built to a much higher standard of thermal thresholds. The 2nd reason is that traditional cooling control mechanisms and policies employed in datacenters are based on a limited number of distributed physical sensors on the ceiling of the cold and hot aisles. These sensor measurements cannot reflect the overall picture of a fluctuating temperature field. The non-uniform temperature distribution at server inlets caused by air flow patterns, recirculation or bypass air between hot and cold aisles, etc. won't be picked up from ceiling sensors. Given the limitation of using just a few sensors to determine ambient temperature over the large and varied physical space of a data center, low temperature traditional air cooling was required. The 3rd reason is the indirect connection between demand (server) and supply (CRAC) in the traditional control policy. Different kinds of servers with different power densities and power distribution always means different cooling requirements (power variation, inlet air temperature, airflow volume). A CRAC unit cannot know and reflect the exact cooling demand directly based on changing server usage. For example, a multi-server application workload increases causing a large power variation and subsequent increased thermal output. The traditional cooling control mechanism will not response until the physical ceiling sensor notices the in-room air temperature rise and reports this to the BMS system. This longer response time causes a larger temperature oscillation range, and therefore a conservative operational temperature setting is needed to ensure the thermal safety of the DC equipment. Higher operational temperatures reduces that thermal headroom for equipment as well as reduces allowable temperature variations in an IDC. IDC operators have less response time to deal with any large power/thermal variation or IDC cooling infrastructure in the event of a failure. To overcome these IDC limitations and successfully move to an HTA state, a customized design, both for the cooling infrastructure and server infrastructure is required. An important trend in the industry called ODC (On Demand Cooling) with real time thermal controller provides us an opportunity to fix the above problems. ODC is a new solution for IDC designers to improve cooling responsiveness. This cooling solution looks at the server and cooling infrastructure together from the very beginning of a concept design phase. The connection between demand and supply is now integrated, and new DC technologies are being developed to optimize these designs. An example of ODC is taking server inlet temperature and air volume flow rate demand of servers and using that in realtime to control the supply air temperature and supply airflow volume to accurately meet demand. Power threshold data can also be used in the same way. Real-time server power metrics can be used to alert when thermal output is increasing and avoid a potential temperature spike. Using ODC control with real time thermal information, reduces temperature oscillation and provides a quicker response from thermal issues. By establishing a smart thermal operation with characteristics like direct feedback, accurate control and quick response, HTA can safely be achieved with confidence. We will explore implementing an aggressive thermal management control policy with thermal safety features. The whole process of implementing an ODC infrastructure, including data collection and data transfer/data processing is explained in section 2. The control policy of ODC will be described. Key indexes defined to judge the improvement and progress of ODC control methods, compared with traditional ones. A case study is documented showing how to implement ODC in a specific MDC module. Important components of the tested ODC design are temperature field distribution, response time, response quality (temperature oscillation) in specific situations (i.e. power variation), non-uniform power distribution, and CRAC failure analysis. The results from testing show that the ODC, with real time thermal information, can reduce temperature oscillation and response time when dealing with the fluctuating temperature conditions. ODC shows it can support raising the operational temperature in an IDC and can help save significant cooling energy.