MAnagement of DYnamic NEtworks & Services 2003-2012 Networking & Software Security Theme within LORIA Networking (1B) Theme at l INRIA Slides v3.0, based on Manuscript v3.0 of July 24th, 2003 Members Challenges Research activities Madynes dans «Observation et supervision de systèmes complexes, répartis r et dynamiques 23 Septembre 2003-1 -
Participants (10/2003) Scientific Leader Olivier Festor [INRIA Researcher, Hab.] Project officer Isabelle Chrisment [AP ESIAL-UHP-Nancy 1] Group members Laurent Andrey [AP U.Nancy 2, Dept Manager SRC IUT Verdun] Laurent Ciarletta [AP ENSMN, since 1/10/2003] Jacques Guyard [Professor ESIAL-UHP] Emmanuel Nataf [AP U. Nancy 2] Josiane Reffort [Secretary, LORIA-UHP] André Schaff [Professor, ESIAL-UHP, ESIAL Director] Radu State [INRIA Researcher, since 1/10/2002] Ph.D. Students Mouna Benaissa [10/2003] Hassen Sallay [12/2003] Guillaume Doyen [10/2005] Rémi Badonnel [10/2006] Mohamed Sallah Bouassida [10/2006] Vincent Cridlig [10/2006] Engineers (yearly contracts) Abdelkader Lahmadi [10/2003] 23 Septembre 2003-2 -
Network and Service Management Definition Process(es) of controlling the resources of a complex entity (device, network, service) so as to maximize its efficiency and productivity Basic activities Observe (monitor) Act (provision, configuration) Plan Functional areas Faults Configuration Accounting Performance Security Business objectives (e.g. Gold Service) Services Networks Elements 23 Septembre 2003-3 -
New challenges (1) The need for network and service management grows at a very fast rate due to : Multiplication of services and service offers, Multiplication of access points, Increasing dependency towards services and the underlying infrastructure(s) Growth in quality requirements BUT need to reduce operational and management costs Network and service management becomes more and more complex It already was complex J Multi-domain becomes unavoidable and new models emerge (e.g. trust models, community domains, ) # of components that participate in the service delivery chain are increasing exponentially. 23 Septembre 2003-4 -
New challenges (2) System name status Static Management Manager/MIB/Agent Manual configuration of participating entities Homogeneous management domains Pre-sharing of management information modes Scales Time Jain, J2EE, Few seconds to a few month 5-8 years for information models (standards) Size : 1 to a few hundred Dynamic Management Infrastructures, services, usage & mobility Multiple and shared administrative domains Variable models and management interfaces Negotiable & evolutive management functions Scales Time : ms to a few seconds Size : 1 to several millions/billions Autonomous Management becomes critical! 23 Septembre 2003-5 -
The MADYNES Research Themes We manage it all, but all we do is management! Mobile Self Organizing Networks Configuration & Provisioning P2P Autonomous management Web Services Information Models Self organizing Management Frameworks Multicast Functional areas IPv6 Performance & Availability Monitoring P2P & Application Level Routing Evaluation & Benchmarking Security Management GRID VoIP OSGi Application domains J2EE 23 Septembre 2003-6 -
Autonomous Management Theme Participants L. Andrey, I. Chrisment, G. Doyen, O. Festor, E. Nataf, R. State Addressed problems Master the dynamics of management and managed infrastructure Harmonize the time scales Master heterogeneity Validate & benchmark? 23 Septembre 2003-7 -
Management plane self-configuration & P2P Self-configuration of the management plane Discovery of the management plane Who/where is my manager? Automated configuration of the management plane What shall I do? What information do I need to export? At which frequency? What data is shared among managers? Design and evaluate several management models P2P models Universal naming and identity issues Extended manager/agent paradigm Application level routing Management Information type based Managed object instances based We already used this type of routing in CmisJavaAPI & Open Master 23 Septembre 2003-8 -
Evaluation & benchmarking Management cost models Cost may be neglected in large environments (e.g. J2EE)? Is the 5% limit correct? What about JMX in a J2ME within my PDA? 5-95% overhead (memory, CPU, )? Impact on the service performance? Management activity & monitoring models Get/Design/Elaborate management traffic patterns Impact of management traffic on constrained environments E.g. wireless sensor networks Scalability issues What management functions have to be distributed? Design of service efficient monitoring algorithms Management overhead Service logic Ze Real Internet Topology 64KBps 23 Septembre 2003-9 -
Madynes dans «Observation et supervision de systèmes complexes, répartis et dynamiques» Impact de cette observation? Surcharges (cpu, mémoire, réseau) liée à l observation de la partie «fonctionnelle» du système La mesurer? La limiter? Comment se font les relations entre les composants du système et l administration? Même modèle(s) pour tous les plans? 23 Septembre 2003-10 -
Propositions Bibliographie en cours Travaux de Adbu, Hanan, Bauer «placement des fonctions de gestions» Fonctions de coûts D où viennent-elles? Bench d un middleware de supervision (ex : jmx). En cours (L. Andrey, E. Cecchet) :-/ Est-ce que ce problème à finalement un sens dans une infrastructure «auto-*»? Reste le test du plan de supervision Les composants fonctionnels ont-ils besoin d info venant de la supervision? Infrastructures de test Obtenir une idée des coûts «unitaires» de choses comme : heartbeat, gauge d une valeur (ex : compteur de messages) Comment tout ceci s intègre dans les composants Remarque : nous continuons à coopérer avec FT R&D Grenoble autour de Fractal et administration. Comment un composant perçoit le plan de supervision? Qu est-ce que le plan de supervision peut demander à un composant? Ex : cycle de vie, reconfiguration 23 Septembre 2003-11 -