An Informal Publication from Academia’s Premiere Storage Systems Research Center Devoted to Advancing the State of the Art in Storage and Information

The Data Center Observatory (DCO) was conceived to guide and enable experimental research into data center automation and effi ciency. The DCO will be a fully-instrumented operational data center at Carnegie Mellon, simultaneously used as a utility and as a research vehicle. Its computation and storage resources will be a utility for CMU research activities, from data mining to CAD/architecture to visualization, and real networked services. At the same time, PDL researchers will measure all operational aspects, including power/thermal, human administrator time, resource utilization patterns, failures and their consequences, and so on. Further, technologies and tools developed to mitigate operational costs will be deployed in this real environment, and measurements will quantify their effi cacy and guide refi nement in an interative research cycle. Why a Data Center Observatory is Needed There is broad agreement that human administration and other operational costs are major issues for data centers—perhaps the biggest challenge facing modern computing and certainly the dominant factor in total cost of ownership. But, there is little specifi c information about where administration time and effort goes and the breakdown of operational costs. As a consequence, it is diffi cult to focus researchers and engineers on the real problems—it is diffi cult to solve a problem that cannot be specifi ed. Further, for many known administration challenges, such as diagnosis of performance problems or application failures, it is diffi cult to evaluate the effectiveness of a proposed solution. For better or worse, “administratability” does not appear to lend itself to controlled lab experiments. The data center observatory was conceived as an aggressive, but necessary, case study for understanding operational costs and evaluating solutions. Only by measuring and characterizing operational costs in real environments can we focus research attention on the most critical unsolved aspects. Explicitly instrumented to act as an observatory, the DCO will be one such real environment. We hope to deploy proven instrumentation technologies,