Power Consumption Analysis of Replicated Virtual Applications in Heterogeneous Architectures
Nowadays, power consumption in IT infrastructures is a major area of concern for both academia and industry. In data centers where computational power is provided by means of virtualized resources, like virtual machines, the policy to allocate them on phy
- PDF / 336,598 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 26 Downloads / 184 Views
Abstract Nowadays, power consumption in IT infrastructures is a major area of concern for both academia and industry. In data centers where computational power is provided by means of virtualized resources, like virtual machines, the policy to allocate them on physical servers can strongly impact the power consumption of the entire system. This affects data center management, and proper estimation means can offer an important guidance to administrators. We propose a lumped Petri net model to investigate the contribution to energy efficiency due to different allocation and deallocation policies on heterogeneous machines with different power demands, to support estimation and planning of datacenter needs.
Keywords Energy efficiency Generalized stochastic petrinets acenters Allocation policies Performance evaluation
Virtualized dat-
G. Ciardo A. Miner Department of Computer Science, Iowa State University, 226 Atanasoff Hall, 50011 Ames, USA e-mail: [email protected] A. Miner e-mail: [email protected] M. Gribaudo (&) P. Piazzolla Dipartimento Di Elettronica, Informazione E Bioingegneria, Politecnico Di Milano, Via Ponzio 34/5, 20133 Milano, Italy e-mail: [email protected] P. Piazzolla e-mail: [email protected] M. Iacono Dipartimento Di Scienze Politiche, Seconda Università Degli Studi Di Napoli, Viale Ellittico 31, 81100 Caserta, Italy © Springer International Publishing Switzerland 2016 L. Caporarello et al. (eds.), Digitally Supported Innovation, Lecture Notes in Information Systems and Organisation 18, DOI 10.1007/978-3-319-40265-9_21
285
286
G. Ciardo et al.
1 Introduction The commercial scenario of IT in the current decade is dominated by a constant demand for intelligent services and the availability of increasingly massive volumes of data. These two forces drive the market towards a consolidation of complex software systems into large data centers that host evolving, heterogeneous, high– performance computing architectures. The aspects related to computing, networking, and storage are very complex, as a consequence of the scale, the interrelation, the volume of data, the reliability requirements, and the diversity in workload. The authors already investigated these aspects in [1–4, 7, 8, 16]. However, there are other reasons for which the management of such infrastructures is a challenging problem: besides issues strictly connected to computing and networking, energy problems arise for both powering the infrastructure and cooling down its components. Energy requirements represent a significant part of costs and affect the ability of providers to stay in the market (at the point that energy related attacks have been designed to damage providers (see [14, 20])). Virtualization is a key software technology on which proper resource scheduling solutions can be designed: the literature offers many proposals (e.g. see [24, 26]), but the research field is just opening. Recent results on energy issues in these systems are pointed by [18] and related papers. Because large–scale empirical ex
Data Loading...
