Non-Markovian Modeling of a BladeCenter Chassis Midplane Salvatore Distefano 1 , Francesco Longo 2 , Marco Scarpa 2 , and Kishor S. Trivedi 3 1 Dipartimento di Elettronica, Informazione e Bioingegneria, Politecnico di Milano, 20133 Milano, Italy salvatore.distefano@polimi.it 2 Dipartimento di Ingegneria DICIEAMA, Universit`a di Messina, 98166 Messina, Italy {flongo,mscarpa}@unime.it 3 Department of Electrical and Computer Engineering, Duke University, 27708-0291 Durham, USA ktrivedi@duke.edu Abstract. In distributed contexts such as Cloud computing, the relia- bility and availability of the provided resources and services have to be assured in order to meet user requirements. At the infrastructure level, this specification is translated into tighter ones on the datacenter hosting physical resources. In this paper, starting from a real case study of the IBM BladeCenter, we provide a technique for the quantitative evaluation of datacenter infrastructure availability. The proposed technique allows one to take into account both aging phenomena and multiple operating conditions. In particular, one subsystem of the BladeCenter, the chassis midplane, is studied. Indeed, based on the stochastic characterization of the midplane reliability through statistic measurements, a model dealing with the non-exponential failure time distribution thus obtained is evalu- ated to demonstrate the suitability and the effectiveness of the proposed technique. 1 Introduction New technologies and applications strongly impact on everyday life, aiming at improving quality standards. IT infrastructure lies at the heart of such tech- nologies, acting as the engine of the digital “revolution”. Blade server systems are becoming a de-facto standard architecture in distributed computing infras- tructure. Indeed, they are used in many academic and business contexts, such as e-commerce, banking, financial, stock trading, and telephone communications as well as in research applications, in addition to several types of life-critical and safety-critical systems and services. Furthermore, blade server systems are the pillars on which Cloud, Web, and social network technologies are based. A primary requirement a blade server has to fulfill is related to its availability (and associated downtime). A common practice is to consider server modules (blades) as stand-alone servers with shared services rather than the blade server A. Horv´ath and K. Wolter (Eds.): EPEW 2014, LNCS 8721, pp. 255–269, 2014. c  Springer International Publishing Switzerland 2014