Upgrade to Pro — share decks privately, control downloads, hide ads and more …

SRC Workshop - Complexity of Ambient Software

SRC Workshop - Complexity of Ambient Software

Systems Research Challenges Workshop - @RedHat @Newcastle University
http://www.ncl.ac.uk/digitalinstitute/news/item/systemsworkshop.html

2a4bd42b6a3db5858ff2250e236f9631?s=128

Frédéric Le Mouël

January 17, 2017
Tweet

Transcript

  1. Complexity of Ambient Software: from Dynamic Composition to Distributed, Contextual,

    Autonomous, Large-scale Execution January, 17 2017 Frédéric Le Mouël University of Lyon - INSA Lyon @flemouel SRC IoT Workshop
  2. Agenda - Middleware & Ambient Intelligence - Towards Dynamic, Scalable,

    Autonomous Middleware - Middleware Perspectives 2
  3. The Beginning 3 Heterogeneity Single Machine API Hardware Issues Application

    Domain Evolution Software Challenges Multi standards Gateways Internet Providers 1990 Impacts
  4. The Breakthrough 4 Heterogeneity Single Machine API Hardware Issues Application

    Domain Evolution Software Challenges Multi standards Gateways Internet Providers Jini Web Services SOA 2000 REST Impacts
  5. Dynamism 5 Heterogeneity Dynam ism Single Machine M2M API Mobile

    Objects VANET Home Automation Hardware Issues Application Domain BANET Sensors Complexity Evolution Software Challenges Multi standards Gateways Internet Providers Service Composition Context-Oriented Impacts
  6. Scalability 6 Heterogeneity Dynam ism Scalability Single Machine M2M User

    Social Group API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Data Centers Smartphone Fleet Deployment BANET Sensors Complexity Evolution Software Challenges Multi standards Gateways Internet Providers Discovery Cloudlets Message-Oriented Middleware Event-based Processing Impacts
  7. Autonomy 7 Heterogeneity Dynam ism Scalability Autonom y Single Machine

    M2M User Social Group Society API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Big Data Data Centers Smartphone Fleet Deployment Drone Fleets Autonomous Vehicles BANET Sensors Service Robotics Event-based Processing Internet of Things Complexity Evolution Software Challenges Multi standards Gateways Internet Providers Active Assisted Living Message-oriented Middleware Discovery Cloudlets Machine Learning Self-Managed Distributed Systems Impacts
  8. 8 Heterogeneity Dynam ism Scalability Autonom y Single Machine M2M

    User Social Group Society API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Big Data Data Centers Smartphone Fleet Deployment Drone Fleets Autonomous Vehicles BANET Sensors Service Robotics Deep Learning Event-based Processing Internet of Things Self-managed distributed systems Complexity Evolution Software Challenges Multi standards Gateways Internet Providers Active Assisted Living Message-oriented Middleware Discovery Cloudlets 1 2 3 Impacts
  9. 1. How to deal with dynamism? 2. How to overcome

    scalability issues? 3. How to distribute decision-making? Research Contributions 9
  10. 10 Ambient Environment

  11. 11 Execution Environment Heterogenous Not Reliable Mobile

  12. 12 Application Execution Flow

  13. 13 Where? Application Ambient Environment

  14. 14 What? Fine-grained Method-level Coarse-grained Bundle & Service-level [Golchay 2016]

    [Ibrahim 2008, Ben Hamida 2010]
  15. 15 When? Automatic Proximity Cloud Spontaneous Semantic Service Composition [Golchay

    2016] [Ibrahim 2008]
  16. 16 How? Graph Cut with Multiple Destinations Collaborative Decision Cache

    Graph Coloring ACO Algorithms [Golchay 2016] [Ben Hamida 2010, Golchay 2016]
  17. 17 ~10 devices ~50-100 services [Ben Hamida 2010] [Golchay 2016]

    [Ibrahim 2008] Time (ms) Node Number Execution Time AxSel with adaptation AxSel without adaptation Efficient ~25-55ms
  18. Guidelines - Dynamism - Engineering Granularity - good offloading performance

    - Environment Volatility - good reactivity - Service & Semantics - bad scalability? 18
  19. 1. How to deal with dynamism? 2. How to overcome

    scalability issues? 3. How to distribute decision-making? Research Contributions 19
  20. 20 Subscriber S1 channel.subscribe( (topic = « Temperature », value

    = [25,35]), (topic = « Location », value = [7,13]) ) Publisher P1 channel.publish( (topic = « Temperature », value = 28), (topic = « Location », value = 12) ) Matching Time?
  21. 21 Subscribers Topic Ranges Geometric Point Enclosure Problem

  22. 22 Index Structure: H-Tree - Interval-based - Tagging non-matching subscribers

    REIN
  23. 23

  24. 24 Pri-REIN S3 will be served before S1 + Matching

    Time intervals
  25. 25

  26. Guidelines - Scalability - Message-oriented Middleware - asynchronous - Distributed

    Publish/Subscribe - efficient, QoS - Engineering - genericity - Content Relevancy? 26
  27. 1. How to deal with dynamism? 2. How to overcome

    scalability issues? 3. How to distribute decision-making? Research Contributions 27
  28. 28 Short-Path Problem Guidance Service Fuel Consumption Travel / Waiting

    Time Optimization
  29. 29 Local vs Global Optimization Decision Partial Data

  30. 30 When moving & service connected, what data to exchange?

    Path modification decision?
  31. Ant-inspired Distributed Decision-making 31 ACO (Ant Colony Optimization) Vehicle Ant

    Pheromone Evaporation
  32. 32 Data exchange: Pheromone map of vehicle m : Travel

    time at the maximum allowed speed Travel time measured by m at time t 0 The more is high, the more the information is old Pheromone evaporation: Pheromone validity time Evaporation gradient 0.5 init for unknown places
  33. PKP KPP PDLAIS PPE CS 11,6 % 1,8 % 3,7

    % 7,9 % 3,7 % Travel Time Gain k-path without pheromone k-path with pheromone Autonomous Intersections Local Pheromone Centralized Solution Normal Traffic 33
  34. 34 k-path without pheromone k-path with pheromone Autonomous Intersections Local

    Pheromone Centralized Solution Earthquake PKP KPP PDLAIS PPE 80 % 40 % 20 % 20 % Arrival Percentage PKP KPP PDLAIS PPE CS 98 % 85 % 78 % 89 % 78 % Accident Arrival Percentage
  35. Guidelines - Autonomy - Local decisions - can be globally

    efficient - Local decisions - robustness - Greatly depends on the use-case - Smart City: traffic ≠ parking 35 Tradeoff in favor of local decisions [Lin 2015, Lèbre 2016]
  36. Concluding Remarks - Technology is here! - Middleware Dynamism, Scalability,

    ok! - Smart Middleware: Natural Receptacle for Autonomy! - Engineering 36
  37. Concluding Remarks - Why are not Middleware & Ambient Intelligence

    in production ? - (when Middleware & Cloud Computing are main trend!) 37 & Internet of Things & Vehicular Networks
  38. 38 Heterogeneity Dynam ism Scalability Autonom y Single Machine M2M

    User Social Group Society API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Big Data Data Centers Smartphone Fleet Deployment Drone Fleets Autonomous Vehicles BANET Sensors Service Robotics Deep Learning Event-based Processing Internet of Things Self-managed distributed systems Software Challenges Multi standards Gateways Internet Providers Active Assisted Living Message-oriented Middleware Discovery Cloudlets Complexity Evolution 2000 Impacts
  39. 39 Heterogeneity Dynam ism Scalability Autonom y Single Machine M2M

    User Social Group Society API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Big Data Data Centers Smartphone Fleet Deployment Drone Fleets Autonomous Vehicles BANET Sensors Service Robotics Deep Learning Event-based Processing Internet of Things Self-managed distributed systems Software Challenges Multi standards Gateways Internet Providers Active Assisted Living Message-oriented Middleware Discovery Cloudlets Complexity Evolution 2006 2007 iPhone Facebook Impacts
  40. Perspectives - User & Society acceptance ↗ - Hot Research

    Issues: - IoT Security - IoT Automatic Provisioning & Deployment - IoT Safety with Distributed Behavior Checking 40
  41. Perspectives - Planetary-scale Middleware & Distributed Systems - Interconnecting Smart

    Cities - Internet of People 41 Birds W ater Understanding Earth Macro-behavior Distributed really anywhere
  42. Future 42 Heterogeneity Dynam ism Scalability Autonom y Single Machine

    M2M User Social Group Society API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Big Data Data Centers Smartphone Fleet Deployment Drone Fleets Autonomous Vehicles BANET Sensors Service Robotics Deep Learning Event-based Processing Internet of Things Self-managed distributed systems Software Challenges Multi standards Gateways Internet Providers Active Assisted Living Message-oriented Middleware Discovery Cloudlets Complexity Evolution Impacts
  43. Future 43 Heterogeneity Dynam ism Scalability Autonom y Single Machine

    M2M User Social Group Society API Mobile Objects Context-oriented VANET Home Automation Hardware Issues Application Domain CRAN Cloud Computing Big Data Data Centers Smartphone Fleet Deployment Drone Fleets Autonomous Vehicles BANET Sensors Service Robotics Deep Learning Event-based Processing Internet of Things Self-managed distributed systems Software Challenges Ethics Humanity Privacy by design Affective Computing Neural Connectivity Human Enhancements Quantum Computers Avatars Augmented Reality Nano Robots Smart Dust Multi standards Gateways Internet Providers Active Assisted Living Message-oriented Middleware Discovery Cloudlets Complexity Evolution Ethical Software Life-cycle Impacts
  44. Questions? frederic.le-mouel@insa-lyon.fr @flemouel http://www.le-mouel.net http://dynamid.citi-lab.fr 44

  45. Bibliography [Ibrahim 2008] N. Ibrahim, Spontaneous Integration of Services in

    Pervasive Environments, PhD Thesis, INSA Lyon, Lyon, France, September 2008. [Ben Hamida 2010] A. Ben Hamida, AxSeL : un intergiciel pour le déploiement contextuel et autonome de services dans les environnements pervasifs, PhD Thesis, INSA Lyon and ENSI, University of La Manouba, Lyon, France, February 2010. [Qian 2015] S. Qian, J. Cao, F. Le Mouël, M. Li, and J. Wang, Towards Prioritized Event Matching in a Content-based Publish/Subscribe System. In Proceedings of the 9th ACM International Conference on Distributed Event-Based Systems (DEBS'2015), pp. 116–127, Oslo, Norway, June 2015. [Lin 2015] T. Lin, Smart Parking : Network, Infrastructure and Urban Service, PhD Thesis, University of Lyon, INSA Lyon, Lyon, France, December 2015. [Golchay 2016] R. Golchay, From Mobile to Cloud : Using Bio-Inspired Algorithms for Collaborative Application Offloading, PhD Thesis, University of Lyon, INSA Lyon, Lyon, France, January 2016. [Lèbre 2016] Marie-Angle Lèbre, De l’impact d’une décision locale et autonome sur les systèmes de transport intelligent à différentes échelles, PhD Thesis, University of Lyon, INSA Lyon, Lyon, France, January 2016. [Le Mouël 2016] Frédéric Le Mouël, Complexité du logiciel ambient : de la composition dynamique à l’exécution distribuée, contextuelle, autonome et large-échelle, Habilitation Thesis, University of Lyon, INSA Lyon, Lyon, France, November 2016. 45 — “Family” Extract of “Ellyn’s Elements of Style” 07/08/2010