Received: 15 December 2017 Revised: 11 April 2018 Accepted: 13 April 2018 DOI: 10.1002/ett.3425 SPECIAL ISSUE ARTICLE Balancing QoS and power consumption in programmable 5G infrastructures Alessandro Carrega Giancarlo Portomauro Matteo Repetto Giorgio Robino S3ITI National Lab, CNIT, Genova, Italy Correspondence Matteo Repetto, S3ITI National Lab, CNIT, 16145 Genova, Italy. Email: matteo.repetto@cnit.it Funding information European Commission H2020 under the projects ARCADIA and MATILDA, Grant/Award Number: 645372 and 761898 Abstract The advent of fifth generation is promising to push far more intelligence than today to the network boundary, hence boosting novel computing models based on fog/edge paradigms. The need for proximity in computation, coupled with various forms of mobility, will be responsible for dynamic shifting of workload within the system, with large fluctuations in resource usage. This eventually turns into poor energy efficiency of the whole infrastructure. However, improv- ing efficiency usually deteriorates quality of service, hence the dilemma about how to balance these two contrasting aspects. In this paper, we propose a framework that leverages the increasing pro- grammability of ICT infrastructures to pursue a linear relationship between power consumption and workload, while safeguarding quality of service. Our approach is based on workload consolidation and extensions to existing cloud management software. We collected both real measurements from an exper- imental testbed and performance analysis from simulations to evaluate the consolidation strategy in more complex environments. 1 INTRODUCTION The underpinning concept behind fifth generation (5G) is a radical transformation of communication networks, ie, beyond mere performance increase (latency, bandwidth, reliability, and density), 5G envisions the installation of capillary computing and storage resources. All this infrastructure builds a rich and pervasive information and communications technology (ICT) fabric, which turns networks into intelligent orchestration platforms. 1,2 Different from the cloud, where resources are located in one or a few big data centers, 5G infrastructures will consist of large distributed, pervasive, het- erogeneous, and multidomain environments, with both traditional data centers and a large number of small/tiny edge installations. This ultimately complicates management and poses new challenges for deployment and orchestration. The 5G infrastructures will be specifically designed to fulfill the challenging requirements of many vertical industries, ie, smart manufacturing, energy, e-health, automotive, media, and entertainment. 2 With the grow of user-centric services, characterized by interactivity and strict constraints on latency, the workload is expected to follow user distribution and mobility patterns, hence leading to uneven, unsteady, and nonuniform distribution within the network. Consequently, the usage of peripheral installations is expected to dynamically vary with hourly, daily, weekly, and seasonal periodicity. According to current management practice in telecommunication networks, the whole system is expected to be largely overdimensioned. In fact, breaking ongoing services is considered a severe quality detriment, and it is far less tolerable than denying set-up of new ones. This immediately turns into low resource usage and large energy waste, which means an unfair and inefficient balance between the two quantities. Trans Emerging Tel Tech. 2018;e3425. wileyonlinelibrary.com/journal/ett Copyright © 2018 John Wiley & Sons, Ltd. 1 of 18 https://doi.org/10.1002/ett.3425