On Server and Path Selection Algorithms and Policies in a light Content-Aware Networking Architecture Eugen Borcoci, Marius Vochin, Mihai Constantinescu, University POLITEHNICA of Bucharest Bucharest, Romania emails: eugen.borcoci@elcom.pub.ro, marius.vochin@elcom.pub.ro, mihai.constantinescu@elcom.pub.ro Jordi Mongay Batalla, Warsaw University of Technology / National Institute of Telecommunications Warsaw, Poland jordim@interfree.it Daniel Negru, LaBRI Lab, University of Bordeaux, Bordeaux, France daniel.negru@labri.fr Abstract —Appropriate content server and path selection procedures based on different algorithms constitute the first set of actions to be performed in content delivery systems. Multi-criteria optimization algorithms based on user context, network and servers information can be used to enhance the overall efficiency. This paper contains a preliminary work, focused on algorithms and policies for optimized paths and server selection, aiming to finally implement a subsystem in the framework of a content delivery light architecture system. Keywords — Content delivery, Server selection, Path selection, Content-Aware Networking, Multi-criteria decision algorithms, Future Internet. I. INTRODUCTION The content orientation is an important trend recognized in the current and Future Internet [1] The Information/Content-Centric Networking (ICN/CCN), approach [2][3], revisits some main concepts of the architectural TCP/IP stack. In parallel, “light ICN”. evolutionary solutions introduce Content-Awareness at Network layer (CAN) [4]. Seen partially as an orthogonal solution, Content Delivery Networks (CDNs) improve the content services [5] by distributing the content replica to cache servers located close to groups of users. However, the above solutions involve complex architectures, high CAPEX and significant modifications in Service/Content Providers and Network Providers/Operators. The DISEDAN Chist-Era project [6][7], (service and user-based DIstributed SElection of content streaming source and Dual AdaptatioN, 2014-2015) proposes an evolutionary and light architecture to enhance the content delivery via Internet. It studies pragmatic solutions for the multi-criteria hard problem of best content source selection, considering user context, servers availability and possibly network status information (if available). The novel concept is based on: a. two-step server selection mechanism (at Service Provider (SP) and at End User) by using algorithms that consider context- and content-awareness; b. dual adaptation mechanism during the sessions, consisting of media flow adaptation and/or content servers handover. The solution could be rapidly deployed in the market since it does not require complex architecture like ICN, full-CAN or CDN. This paper contains a preliminary work on paths and server combined selection algorithms and policies applicable by SPs in a light content delivery architecture Section II is a short overview of related work. Section III outlines the overall system and problem description. Section IV contains the main paper contributions, focused on: a. paths and content server selection combined algorithm; b. modifications to allow introduction of SP policies, aiming to increase the system flexibility. Section V contains conclusions and future work outline. II. RELATED WORK ON MULTI-CRITERIA DECISION ALGORITHMS This section is a very short overview on some previous work related to path-server selection in content delivery systems, based on Multi-Criteria Decision Algorithms (MCDA). The problem belongs to the more general one known as multi-objective optimization. This has been extensively studied in various and large contexts of economics and engineering. The paper will not detail this. Few references are given at the end of the paper [8][9][12]. The general problem of multi-objective optimization is to find min F(x) = [f 1 (x), ..f k (x)] where x  X i , the decision variables space, and f 1 (x), ..f k (x), are a set of objectives, [8] [9]. Such problems are in general NP complete, so, different simplified heuristics have been searched. A simple scalar approach maps the k-dimensional vector onto a single scalar value w by using an appropriate cost function c(), thus reducing the problem to a single-criterion one. However, information about individual components is lost. In the server-path selection problem, several decision parameters are important, such as: server load and proximity, transport path (length, bandwidth, loss, and jitter). Solutions have been searched treating the decision variables separately and considering them as independent. Note that in our case this is only partially true, e.g., delay and jitter are clearly not independent variables. Therefore modifications should be added to the basic algorithm to capture such effects and this paper proposes a solution. The reference level decision algorithms, [10][11], considers a decision space R m and the decision parameter/variables: v i , i=1, ..m; i, v i ≥0. A candidate solution is an element S s =(v s1 , v s2 , .., v sm )  R m . Let S be the number of candidates indexed by s = 1, 2, ..S. The value ranges of decision variables might be bounded by given 17 Copyright (c) IARIA, 2014. ISBN: 978-1-61208-368-1 ICSNC 2014 : The Ninth International Conference on Systems and Networks Communications