Proactive Workload Dispatching on the EURORA Supercomputer Andrea Bartolini, Andrea Borghesi, Thomas Bridi, Michele Lombardi, and Michela Milano DISI, University of Bologna, Italy {a.bartolini,andrea.borghesi3,michele.lombardi2,michela.milano}@unibo.it thomas.bridi@gmail.com Abstract. In the era of Cloud Computing, Big Data, and Quantum Physics Simulations, data centers play in the world ICT infrastructure a role as big as (sadly) their power consumption. In many cases, a surpris- ing amount of such consumption is due to idle resources, either intro- duced to face workload peaks or leftovers of workload fragmentation. In this context, proactive workload dispatching has the chance to improve the utilization of computing resources, thus reducing the idle time and improving the ability to handle peaks. In this paper, we devise a CP based approach for proactive workload dispatching on the EURORA su- percomputer placed at the CINECA computing center in Bologna. The new system is evaluated on simulated job traces, where it leads to re- markable improvements in terms of both machine utilization and waiting times for queued jobs with respect to the currently used dispatcher, i.e., Portable Batch System (PBS). 1 Introduction Computing centers play a key role in modern ICT architectures: they run our internet services, keep track of our savings, make our research possible. They are also well known to be power hungry: in Italy, data centers make for ∼2% of the national energy consumption, for a total of 6.6 TWh (roughly that of the Calabria region, according to data by Fondazione Politecnico di Milano, 2010). The mainstream solution to reduce such a gigantic consumption is to em- ploy efficient hardware or efficient design. By doing so, it is possible to obtain remarkable reductions of the PUE index (Power Usage Effectiveness), i.e. the ratio between the power consumption of the whole data center and the power consumption of the IT equipment alone. Recently, a joint effort by the CINECA inter-university consortium [1] in Italy and the Eurotech group [2] has led to the design of the EURORA system. Thanks to an innovative liquid based cooling system and carefully chosen hardware components, this new machine has a PUE of just 1.05 and managed to reach the top of the Green 500 ranking in the first half of 2013, effectively becoming for a time the most efficient supercomputer on earth. As a comparison, PUE values of around 3 were still common in 2009. B. O’Sullivan (Ed.): CP 2014, LNCS 8656, pp. 765–780, 2014. c  Springer International Publishing Switzerland 2014