Laboratoire d’analyse et d’architecture des systèmes
T.GUEROUT, P.LOPEZ, T.MONTEIL, C.ARTIGUES, Y.GAOUA, G.DA COSTA
SARA, ROC, IRIT-UPS
Revue Scientifique : Future Generation Computer Systems, Vol.71, pp.1-17, Juin 2017 , N° 17006
The analysis of the Quality of Service (QoS) level in a Cloud Computing environment becomes an attractive research domain as the utilization rate is daily higher and higher. Its management has a huge impact on the performance of both services and global Cloud infrastructures. Thus, in order to nd a good trade-off, a Cloud provider has to take into account many QoS objectives, and also the manner to optimize them during the virtual machines allocation process. To tackle this complex challenge, this article proposed a multiobjective optimization of four relevant Cloud QoS objectives, using two different optimization methods: a Genetic Algorithm (GA) and a Mixed Integer Linear Programming (MILP) approach. The complexity of the virtual machine allocation problem is increased by the modeling of Dynamic Voltage and Frequency Scaling (DVFS) for energy saving on hosts. A global mixed-integer non linear programming formulation is presented and a MILP formulation is derived by linearization. A heuristic decomposition method, which uses the MILP to optimize intermediate objectives, is proposed. Numerous experimental results show the complementarity of the two heuristics to obtain various trade-offs between the different QoS objectives.
INRIA Sophia, SARA
Rapport LAAS N°17108, Avril 2017, 5p.
Consider a single server queue serving a multiclass population. Some popular scheduling policies for such a system (and of interest in this paper) are the discriminatory processor sharing (DPS), discriminatory random order service (DROS), generalized processor sharing (GPS) and weighted fair queueing (WFQ). The aim of this paper is to show a certain equivalence between these scheduling policies for the special case when the multiclass population have identical and exponential service requirements. In fact, we show the equivalence between two broader classes of policies that generalize the above mentioned four policies. We specifically show that the sojourn time distribution for a customer of a particular class in a system with the DPS (GPS) scheduling policy is a constant multiple of the waiting time distribution of a customer of the same class in a system with the DROS (respectively WFQ) policy.
U.AYESTA, P.JACKO, V.NOVAK
SARA, Lancaster Univ., CERGE-EI, Prague
Revue Scientifique : Journal of Scheduling, Vol.20, N°2, pp.129-145, Avril 2017, doi 10.1007/s10951-015-0456-7 , N° 15412
Many real-world situations involve queueing systems in which customers may abandon if service does not start sufficiently quickly. We study a comprehensive model of multi-class queue scheduling accounting for customer abandonment, with the objective of minimizing the total discounted or time-average sum of linear waiting costs, completion rewards, and abandonment penalties of customers in the system. We assume the service times and abandoning times are exponentially distributed. We solve analytically the case in which there is one server and there are one or two customers in the system and obtain an optimal policy. For the general case, we use the framework of restless bandits to analytically design a novel simple index rule with a natural interpretation. We show that the proposed rule achieves near-optimal or asymptotically optimal performance both in single- and multi-server cases, both in overload and underload regimes, and both in idling and non-idling systems.
A.SIMO TEGUEU, S.ABDELLATIF, T.VILLEMUR, P.BERTHOU
Rapport LAAS N°17083, Avril 2017, 9p.
Z.LIU, D.DRAGOMIRESCU, G.DA COSTA, T.MONTEIL
IRIT, MINC, IRIT-UPS, SARA
Manifestation avec acte : International Conference on Internet of Things, Data and Cloud Computing ( ICC ) 2017 du 22 mars au 23 mars 2017, Cambridge (UK), Mars 2017, 7p. , N° 17076
E.FKI, S.TAZI, K.DRIRA
Revue Scientifique : Future Generation Computer Systems, Vol.68, pp.376-390, Mars 2017 , N° 16468
In recent years, the composition of loosely coupled services with the aim of satisfying the user intention is a widely followed research topic. The composition of services implies the ability to select, coordinate, interact, and interoperate existing services. This is considered as a complex task. This complexity is mainly due to the large number of available services and their heterogeneity as they are created by different organizations. This complexity is increased when services must be dynamically and automatically composed to meet requirements which are not satisfied by existing services. In fact, an approach for service composition must offer the potential to achieve flexible and adaptable applications, by selecting and combining services based of the request and the context of the user. In this perspective, different approaches have been developed for services composition. However, most of the existing composition approaches tend to be static and not flexible in the sense that they do not have the ability to adapt to user requirements. To overcome these challenges, we propose a composition approach in which the generation of the composition schema is performed at runtime through the use of abstract services provided at design time. The composition process that we propose takes as input a structure of user requirements materialized by a graph of intentions and enriches this graph to explicit the implicit relationships. The enriched graph is used to generate an initial composition schema by building the control flow and selecting the appropriate abstract services. The selection of these services is based on the semantic matching and the degree of semantic affinity between abstract services. Then, the final composition schema is generated using a refinement mechanism of abstract services using semantic matching techniques and taking into account user context and constraints.
T.SEREGINA, O.BRUN, R.ELAZOUZI, B.PRABHU
SARA, LIA Avignon
Revue Scientifique : IEEE Transactions on Mobile Computing, Vol.16, N°2, pp.453-465, Février 2017, doi 10.1109/TMC.2016.2546910 , N° 14418
A central problem in Delay Tolerant Networks (DTNs) is to persuade mobile nodes to participate in relaying messages. Indeed, the delivery of a message incurs a certain number of costs for a relay. We consider a two- hop DTN in which a source node, wanting to get its message across to the destination as fast as possible, promises each relay it meets a reward. This reward is the minimum amount that offsets the expected delivery cost, as estimated by the relay from the information given by the source (number of existing copies of the message, age of these copies). A reward is given only to the relay that is the first one to deliver the message to the destination. We show that under fairly weak assumptions the expected reward the source pays remains the same irrespective of the information it conveys, provided that the type of information does not vary dynamically over time. On the other hand, the source can gain by adapting the information it conveys to a meeting relay. For the particular cases of two relays or exponentially distributed inter-contact times, we give some structural results of the optimal adaptive policy.
A.GASSARA, I.BOUASSIDA, M.JMAIEL, K.DRIRA
ReDCAD Laboratory, SARA
Revue Scientifique : Computers and Electrical Engineering, Vol.58, pp.113-125, Février 2017 , N° 17033
In this paper, we present a multi-scale modeling methodology for software System of Systems (SoS) using the formal technique of Bigraphical Reactive System. This methodology provides a correct by design approach ensuring the correctness of the SoS architectures. A first scale is defined by the designer. Then, it is refined by successively adding lower scale details. The transition between scales is implemented following a rule-oriented refinement process. The executed rules respect the system constraints ensuring, in this way, the correctness of the obtained scale architectures. Moreover, we address the dynamic aspect of SoS by providing model-based rules of reconfiguration actions. We illustrate our approach with a Smart Buildings case study.
C.DIETZEL, M.ABT, M.CHIESA, P.OWEZARSKI
DE-CIX, UCL, SARA
Rapport de Contrat : H2020-ICT-2014-1 Project No. 644960, Janvier 2017, 16p. , N° 17007
Over the course of the ENDEAVOUR project the consortium developed a wide range of use cases as potential candidates to be implemented within the ENDEAVOUR prototype. After consolidating the most compelling use cases we implemented them into the ENDEAVOUR platform. To this end, the present deliverable showcases the implemented use cases of the ENDEAV- OUR platform for Internet eXchange Point (IXP) members. Each use case is demonstrated in a video. In addition, Deliverable 4.4 discusses the relevant use cases for IXP operators. In combination, these two deliverables reflect the current state of the ENDEAVOUR platform prototype. We present technical background necessary to understand the implementation of each use case, the high level implementation itself, as well as a workflow of each demonstration.
J.DROMARD, P.OWEZARSKI, M.A.LOPEZ, M.A.MONJAS, A.BASCUNANA, V.MARTIN, F.ARIAS, A.MOZO, S.GOMEZ, B.ORDOZGOITI
SARA, SATEC, ERICSSON, EMC2, UPM
Rapport de Contrat : ONTIC Project (GA number 619633), Janvier 2017, 37p. , N° 17043
Deliverable D5.3 purpose is to provide information about how the algorithms developed in scientific work packages have been applied in use cases. Although it is possible to assume that integration should have been straightforward, in general, adaptations, configurations, and transformations are needed. For instance, the following adaptation could have been needed: Interface adaptation: it means not only protocol (many times a specific protocol wrapper has been designed), but also data model adaptation. Sometimes it has been also interconnected using other off-the-self systems such as data brokers, cloud platforms... Redesign if the algorithm was designed in a language/technology different from the one used in the use case. Also, the following information would be needed in order to fully understand how the algorithms are run: The parameters used in the algorithm implementation within the use case. For instance, if considering a Spark Streaming-based algorithm, the size of the windows or the thresholds used. The configuration parameters for the platform where the algorithm is run (RAM, processors...)