Distributed Models and Algorithms for Telecommunication Systems Management
Explore the principles and applications of distributed models and algorithms for managing telecommunication systems. Topics include fault management, deployment, QoS, service orchestrations, handling workflows, and more. Dive into the foundations for autonomic management, distributed algorithms, and wide-area systems modeling. Discover key figures and the evolution of research in this field.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
DistribCom Distributed Models and Algorithms for the Management of Telecommunication Systems Albert Benveniste, Claude Jard 21 March 2012 Albert Benveniste
Overall Objectives and Topics 21 March 2012 Albert Benveniste
Overall Objectives and Topics Applications Distributed Autonomic Management of network and services Observing: fault management Planning: deployment and reconfiguration Composite Web services emphasis on QoS QoS aware management of service orchestrations Handling Workflows & Data jointly - 3 Albert Benveniste 21 March 2012
Overall Objectives and Topics Foundations for Applications Autonomic Management of Wide Area Distributed Systems Modeling: Petri nets, scenarios Composing: components & interfaces Constructing Models: self-modeling scale-up Distributed Autonomic Management of network and services Observing: fault management Planning: deployment and reconfiguration autonomic Distributed algorithms QoS-aware Management of Wide Area Distributed Systems Probability: uncertainty & non-determinism Performance: time, cost & weights QoS: time, security, availability, reliability, quality Composite Web services emphasis on QoS QoS aware management of service orchestrations Handling Workflows & Data jointly - 4 Albert Benveniste 21 March 2012
People 21 March 2012 Albert Benveniste
History & Flows Axel Legay Fran ois Schwarzentruber Guillaume Aucher Guillaume Aucher F. Schwarzentruber Axel Legay Albert Benveniste Eric Fabre Lo c H lou t Claude Jard Albert Benveniste Eric Fabre Lo c H lou t Claude Jard 2008 2009 2010 2011 Anne Bouillard Blaise Genest Blaise Genest Anne Bouillard New topic: non-classical logics Seeding: statistical model-checking Albert Benveniste 21 March 2012
Background research areas: trans-disciplinarity XXX 50% Albert Benveniste . Models of Concurrency & Formal Methods XXX Eric Fabre. . Logic & Verification XXX Lo c H lou t .. Software Engineering XXX Claude Jard ... Mathematics, Probability & Statistics X Guillaume Aucher Control X Fran ois Schwarzentruber ... Formal Methods in CS + Applied Maths & Control XX Axel Legay .. - 7 Albert Benveniste 21 March 2012
Major results A glimpse of science 21 March 2012 Albert Benveniste
Major results Foundations and Models Efficient and expressive scenario models Modular data structures (trellis unfoldings) Enhanced with Time, Probability, Cost Autonomic management algorithms Autonomic Wavelength Power Tuning in a Photonic Network (demo) Autonomic graceful shutdown and restart Joint network and service active diagnosis in IMS Distributed autonomic algorithms for Diagnosis for Planning Composite Web services & QoS QoS-aware management of service orchestrations (demo) Handling Workflow & Data jointly using Document Based Workflows (demo) Self-modeling Our algorithms are model-based Tens of thousands of components construct models automatically MIBS & Netw discov Standards & Know Models at Need - 9 Albert Benveniste 21 March 2012
ALU-Inria Common Lab Autonomic Wavelength Power Tuning in a Photonic Network limited power regeneration per wavelength limited optical power per fiber Challenges avoid manual tuning & over-dimensioning optimize optical reach without regeneration Distributed and adaptive solution huge coupled non-linear constrained optimization problem distributed P2P tuning with 1 agent per node Autonomic power (re)allocation when connections join or leave Algorithm performs Chaotic Iterations: pick a link at random freeze all optical gains, adjust the gains of the selected link forward the result to the neighbors Results ALU simulations demonstrate 50% reduction in regeneration equipment 2 joint ALU-INRIA patents Albert Benveniste 21 March 2012
ALU-Inria Common Lab Distributed (Factored) Planning Motivation Autonomic cross-domain management of networks & services Moving the system from one state to another state while avoiding some unwanted states and optimizing some cost Hitless maintenance Graceful shutdown & restart Security holes avoidance while reconfiguring system - 11 Albert Benveniste 21 March 2012
ALU-Inria Common Lab Distributed (Factored) Planning Motivation Approach and Result Autonomic cross-domain management of networks & services Optimal planning for a network of interacting automata Moving the system from one state to another state while avoiding some unwanted states and optimizing some cost Features Cross-domain P2P planning Models: local to each domain Distributed Unsupervised: no coordinator Solves the desired optimal planning problem Hitless maintenance Graceful shutdown & restart Security holes avoidance while reconfiguring Pre-requisite Self-modeling shall be applicable system - 12 Albert Benveniste 21 March 2012
ALU-Inria Common Lab Distributed (Factored) Planning: details Planning: drive optimally the network of interacting components to a target state Message Passing Algorithms (MPA) Weighted automata: finite state machine with additive costs on transitions MPA Belief Propagation in Belief Networks algorithm architecture = system architecture Local plans combine into optimal global plan For applications exhibiting local connectivity, factored planning is exponentially faster - 13 Albert Benveniste 21 March 2012
Composite Web services and QoS QoS aware management QoS is multi-dimensional Response time, Throughput Security, Availability, Cost, Orchestrations may not be QoS-Monotonic ( but Br ss) Due to interactions workflow/data/QoS Not noticed in WS community QoS-Monotonicity: new topic Conditions ensuring monotonicity How to deal with lack of monotonicity Albert Benveniste
Composite Web services and QoS GarageB slow GarageB fast QoS aware management CarOnLine request (car) QoS is multi-dimensional Response time, Throughput Security, Availability, Cost, GarageA GarageB timeout no timeout Timeout Timeout mux p Orchestrations may not be QoS-Monotonic ( but Br ss) Due to interactions workflow/data/QoS Not noticed in WS community no yes p=high very slow fast AllCredit AllCredit+ GoldInsure InsureAll InsurePlus min min QoS-Monotonicity: new topic Conditions ensuring monotonicity How to deal with lack of monotonicity c p merge i Orch fast Orch slow CarOnLine response - 15 Albert Benveniste 21 March 2012
Composite Web services and QoS CarOnLine request (car) QoS aware management GarageA GarageB Timeout Timeout mux On top of monotonicity Contract Based Management p no yes car=deluxe Probabilistic Contracts Compare using Stochastic Ordering Monitoring using Page-Hinkley tests Optimal Design and Late Service Binding AllCredit AllCredit+ GoldInsure InsureAll InsurePlus min min c p merge i CarOnLine response QoS calculus (Dioid Algebra) why not a soft bound like this, covering 95% of the cases? Implementation in Orc: weaving QoS in functional spec unfortunately, such contracts do not compose Idea: a contract is a probability distribution - 16 Albert Benveniste 21 March 2012
Composite Web services and QoS Document based workflows In complex business processes: Workflows & Data seen as equal citizens - Workflows can be guarded by document patterns - Workflows can update documents Not possible today: workflows and DB are separate technologies Our objectives Handling Workflow & Data jointly A formal framework for Web-scale distributed service computing Service Interfaces Albert Benveniste
Composite Web services and QoS Document based workflows On top of Active XML documents [Abiteboul] In complex business processes: Workflows & Data seen as equal citizens - Workflows can be guarded by document patterns - Workflows can update documents Not possible today: workflows and DB are separate technologies Our objectives Results A framework for Distributed Document Based Workflows Verification of reachability properties Platform under development Handling Workflow & Data jointly A formal framework for Web-scale distributed service computing Service Interfaces - 18 Albert Benveniste 21 March 2012
Publications See the written document 21 March 2012 Albert Benveniste
Industrial ties transfers & impact 21 March 2012 Albert Benveniste
Industrial ties, transfers & impact Topic Impact ALU-Inria Common Lab Distributed algorithms for the autonomic management Autonomic Wavelength Power Tuning in a ALU-BellLabs Common Lab / HiMa EU IP-Univerself (ALU, Orange) 2 ALU(Inria) patents; detailed exploration of business opportunity by BD, now stalled 1 ALU(Inria) patent Under study mainly with Orange Labs self-modeling Photonic Network Autonomic graceful shutdown and restart Joint network and service active diagnosis in IMS Composite Web services, document based workflows, and QoS Still we think the topic is important So far industrial contact missing (SAP? IBM?) - 21 Albert Benveniste 21 March 2012
Competition & Cooperation 21 March 2012 Albert Benveniste
Competition & Cooperation Topic Community Fundamentals of distributed observation and supervision of wide area distributed systems Formal methods in computer science (AA) Distributed algorithms for the autonomic management of network and services Network & service management . .(A) Distributed systems & algorithms in computer science and control ..(AAA) Web services . ..(A) Data bases .. .(C) Formal methods in ..(AA) Composite Web services, document based workflows, and QoS - 23 Albert Benveniste 21 March 2012
Visibility 21 March 2012 Albert Benveniste
Visibility A. Benveniste is Directeur Scientifique of ALU-BL / Inria common lab A. Benveniste is Directeur Scientifique of Labex CominLabs CominLabs is an Excellence Center (Labex) funded for 10 years Bretagne and Nantes labs in the area of Telecoms & Over-the-top Applications Overall funding of 14M A. Benveniste is member of Orange Labs (and Safran) Scientific Councils E. Fabre is head of the Action de Recherche High Manageability at ALU-BL / Inria common lab L. H lou t is invited researcher in the Indo-French LIA INFORMEL C. Jard is Directeur de la Recherche at ENS Cachan Bretagne - 25 Albert Benveniste 21 March 2012
Future plans DistribCom is not expecting to continue for 4 more years Current plans are to merge DistribCom, S4, and Vertecs teams under the lead of Eric Fabre We present the focus of DistribCom s members in this context 21 March 2012 Albert Benveniste
Seen from our side Some industrial challenges Autonomic management of wide area distributed systems How we d like to contribute (E. Fabre, A. Benveniste) Knowledge plane Knowledge = Models+Data Self-Modeling+Learning Self-Modeling+Learning Cross-domain Cooperative Non-cooperative : Games?? : Distributed optimization Autonomic supervision No central coordinator Many aspects for optimization cost, energy, resilience, security Algorithms supporting multiple QoS dimensions Moving to clouds? Today: centralizing data & management Tomorrow?? - 27 Albert Benveniste 21 March 2012
Seen from our side Some industrial challenges Web-scale Business Processes How we d like to contribute (L. H lou t, C. Jard, A. Benveniste) a and Services Safe and secure Business Processes Pay attention to semantics of distributed programs, have semantically rich interfaces (today syntax only) Multi-dimensional SLA cost, energy, resilience, security Have semantically rich SLA, math models for QoS (not just silver/gold/platinum ) Workflows & Data Document Based + Imperative Workflows Computing as part of composite Services Algorithms as a Service Moving to clouds? Today: centralizing data & management Tomorrow?? - 28 Albert Benveniste 21 March 2012
Future team (this slide is common with the other 3 teams) A new team is under construction (from Vertecs, S4, Distribcom), led by Eric Fabre. Modeling, analysis and management of distributed heterogeneous systems distribution : modularity, composition, concurrency heterogeneity : quantitative aspects, as time, probabilities, costs, performance (QoS), etc analysis : verification, test management : control, diagnosis, planning, optimization... self-modeling : automatic construction of models Challenges scaling up to large&complex by abstractions, approximate analysis, parameterization handling reconfigurable, partially known, open systems designing distributed/modular management methods: modularity, multi-agent, games Applications large open reconfigurable softwares, as web-services or distributed active documents (very) large structured systems: telecommunication network management systems design - 29 Albert Benveniste 21 March 2012
