An evolutionary approach for Cloud learning agents in multi-cloud distributed contexts Antonello Comi ∗ , Lidia Fotia ∗ , Fabrizio Messina † , Giuseppe Pappalardo † ,Domenico Rosaci ∗ , Giuseppe M.L. Sarn´ e ∗ ∗ DIIES Department, University of Reggio Calabria Reggio Calabria, Italy {antonello.comi,lidia.fotia,domenico.rosaci,sarne}@unirc.it † DMI Department, University of Catania Catania, Italy {messina,pappalardo}@dmi.unict.it Abstract—Learning software agents are able to assist Cloud providers in taking decisions about resource management at any level, as they are able to collect knowledge and improve their per- formances over time by means of learning strategies. On the other hand Cloud Federations allow providers to share computational infrastructures in order to build a distributed, interoperable multi-cloud context. In this work we present an evolutionary approach based on agent cloning, i.e. a mechanism of agent reproduction allowing providers to substitute an “unsatisfactory” agent acting in a “cloud context” with a clone of an existing agent having a suitable knowledge and a good reputation in the multi- cloud context. By this approach, cloud agents performances can be improved because they are substituted with agent clones that have shown a better behavior. Keywords—Learning agents; Cloud Computing; XaaS; Proto- col; Cloud Federations. I. I NTRODUCTION Cloud Computing [1] came out with the growing avail- ability of commodity hardware and virtualization technologies, such that computational resources are managed by a suitable software stack [2] relying on virtualization technologies [3]. This technology and usage pattern implied the emergence of the “on-demand computing” business model. As second step, Cloud providers have distributed their data centers around the world, and manage their public cloud data centers in a “collaborative” fashion [4]–[6], [6] When a request holding the specification of certain computing requirements is submitted by a customer it must be fulfilled by determining the (virtual) nodes holding the necessary resources [7], and a suitable location is selected. With the advent of distributed multi-cloud environments, as e.g. Clouds federations [6], [8]–[10], cloud providers can benefit from collaboration and knowledge sharing mecha- nisms. Collaboration is performed by providers whenever XaaS (Anything as a Service) [11] resources are mutually rented to face up shortage of resources. Furthermore, providers of e-services (e.g. SaaS, DaaS) are able to compose complex specialized services by means of single atomic services from several different providers. Knowledge sharing is recognized as a way to improve individual performance of operators which have joined “communities”. This also means making effective use of the knowledge produced by others [12]. The result should be a distributed knowledge management system which allows organizations, at any levels, to take profit from a certain type of collaboration. In such a context, intelligent software agents [13] can assist providers to manage their infrastructures and taking decisions (e.g. resource sharing, composite services, VM migration, and so on). In principle, learning agents are able to operate in environments which are initially unknown, learn knowledge and expertise in a particular area, such that their performance can be improved. This capability is typically derived by the integration of machine learning techniques in multi-agent systems for creating intelligent and adaptive agents [14]. Given the premises above, in this work we propose to adopt an evolutionary approach to allow learning agents to improve their performances over time in a multi-cloud distributed envi- ronment, as Cloud federations [15]. In other words, improve- ments related to learning agents should exploit the advantages to the other agents, and a natural way of obtaining such a result is represented by evolutionary mechanisms [16], [17]. In this approach the concept of evolution, which is a biological concept, is adapted (i.e. simplified) by the following pattern. First of all, a node which is not satisfied by its agent, or that is a newcomer, can require the system to provide in adding or in substitution a new agent. This process happens based on the scores of the learning agents computed by taking into account similarity and reputation in the federation. Whenever any agent is selected to substitute another agent, it is cloned and the created copy is sent to the requester. The remaining of the paper is organized as follows. In Section II we introduce Cloud federations and learning agents. In Section III we discuss the proposed model. Section IV describes the evolutionary approach, while Section V deals with the adopted reputation model. In Section VI we discuss the related work and, finally, in Section VII we present our conclusions and ongoing research. II. LEARNING AGENTS IN MULTI - CLOUD DISTRIBUTED ENVIRONMENTS Cloud providers spread data centers around the world to support users with applications. For small-scale business, private Clouds can assure redundancy and reliability while, on 2015 IEEE 24th International Conference on Enabling Technologies: Infrastructures for Collaborative Enterprises 978-1-4673-7692-1/15 $31.00 © 2015 IEEE DOI 10.1109/WETICE.2015.27 99 2015 IEEE 24th International Conference on Enabling Technologies: Infrastructures for Collaborative Enterprises 978-1-4673-7692-1/15 $31.00 © 2015 IEEE DOI 10.1109/WETICE.2015.27 99 2015 IEEE 24th International Conference on Enabling Technologies: Infrastructure for Collaborative Enterprises 978-1-4673-7692-1/15 $31.00 © 2015 IEEE DOI 10.1109/WETICE.2015.27 99