Distributed Artificial Intelligence with Multi-Agent Systems for MEC
0 likes239 views
The document presents a framework for integrating distributed artificial intelligence through multi-agent systems in Mobile Edge Computing (MEC). It discusses modeling MEC as a multi-agent system, outlining various agent roles and functionalities, and includes a case study of a crowdsensing MEC application. Key challenges addressed are the integration of agent technologies, communication protocols, and ensuring context-aware functionalities within edge systems.
Distributed Artificial Intelligence with Multi-Agent Systems for MEC
- 1. Distributed Artificial Intelligence with Multi-Agent Systems for MEC D.Sc. (Tech.) Teemu Leppänen Center for Ubiquitous Computing, University of Oulu, Finland 1st Edge of Things workshop, ICCCN2019, Valencia, Spain, 1st August 2019
- 2. Outline of the presentation 1. Background – ETSI MEC, software agents 2. Modeling MEC as a multi-agent system 3. Integration of (current) agent technologies into MEC (and edge) 4. Case study: Agent-based crowdsensing MEC application
- 3. Background - ETSI MEC reference architecture • Reference architecture for open multi- vendor edge computing system • Reuses mobile network infrastructure, e.g. base stations and radio network information • Defines edge system components, services, interfaces, KPIs, best practices, … • Design and implementation details omitted • System level: Validation / Resource and application/service lifecycle management • Host level management: Application instantiation, execution and relocation • Challenges: latencies/BW, centralized(?) management, real-time system state, user mobility, …
- 4. Background - Software agents • Classical AI paradigm: Agents are programs that possess capabilities for autonomous operation and decision-making, observe their environment and control their own behavior, actions and interactions. • Reactivity, reasoning, adaptivity, sociality, mobility, planning, learning, proactivity, … • Multi-agent system: Collaborating / cooperating agents solve a problem where the capabilities of a single agent are not enough • Multi-agent systems are one technology for Distributed AI • Well-known agent architectures and framework implementations, e.g. Android • Well-studied interaction protocols, e.g. auctions • ML through reinforcement learning • Main challenge today: How to introduce the agent capabilities, i.e. integrate agent standards and solutions, into IoT and edge computing systems?
- 5. MEC through software agents • We envision Agent-Based Computing as a tool to model, design and implement edge computing systems, while trying to address the complexities • Hierarchical architecture: orchestrator <-> platform <-> host • Distributed architecture: collaboration of components with some autonomy expected in all layers • We see agents as complementary technology with extra capabilities to make edge systems context-aware and less unpredictable • Components implement well-known agents roles • MEC KPIs and APIs provide real-time information to adapt and learn • Challenge: MEC facilitates REST interaction paradigm, how to integrate agent frameworks? 1. Common protocols and proxies/wrappers to translate system component <-> agent interactions 2. REST-compliant agent frameworks
- 6. Agent-based MEC – Roles and functionalities (1/2) • User/developer/stakeholder agents • Represent these as entities in MEC system • Authenticate and negotiate application / resource usage and billing • Manage, collaborate and aggregate in application requests • Represent mobile network operator rules and policies • Orchestration agents (and multi-agent system) • Validate application and service requests • Manage application lifecycles (with stakeholder agents) • Monitor system resource use per service/application • Proactive planning and evaluation of plans for system resource use
- 7. Agent-based MEC – Roles and functionalities (2/2) • Platform management agents (and multi-agent system) • Represent the virtualization infrastructure • Represent hosts on the platform • Manage application lifecycles and platform resource use with orchestration agents, virtualization agents and host agents • Monitor platform resource and virtualization infrastructure use, plan and evaluate • Host management agents • Represent applications and services on the host • Represent virtualization infrastructure on the hosts • Manage application lifecycle on the host and handle data traffic and service requests with other hosts • Monitor host resource use, plan and evaluate
- 8. Case study – MEC-based crowdsensing service 1. MEC service that provides participants for crowdsensing tasks • Uses MEC Location API to follow users across the system 2. MEC application that executes crowdsensing tasks on the system • Based on task requirements (location, data types, movement patterns, etc) receives information on suitable participants from MEC service • Interacts with phone agents (of selected participants) to execute campaigns, based on their requirements and user set constraints 3. User smartphones connected to the MEC system as data sources for applications • Phone agents execute online tasks in the smartphones
- 9. Leppänen, T., Liu, M., Harjula, E., Ramalingam, A., Ylioja, J., Närhi, P., Riekki, J. and Ojala, T. “Mobile Agents for Integration of Internet of Things and Wireless Sensor Networks,” In: IEEE SMC 2013, pp. 14-21, Manchester, UK, 2013. Leppänen, T., Riekki, J., Liu, M., Harjula, E. and Ojala, T. “Mobile Agents-based Smart Objects for the Internet of Things,” In: Fortino and Trunfio (Eds.), Internet of Things based on Smart Objects: Technology, Middleware and Applications, pp. 29-48, Springer, 2014. Leppänen, T., Álvarez Lacasia, J., Tobe, Y., Sezaki, K. and Riekki, J. “Mobile Crowdsensing with Mobile Agents,” Autonomous Agents and Multi-agent Systems, vol. 31, no. 1, pp. 1-35, Springer, 2017. Leppänen, T. Resource-oriented mobile agent and software framework for the Internet of Things. Doctor of Science (Technology) dissertation, C Technica, no. 645, University of Oulu, Finland, 2018. Leppänen, T., Savaglio, C., Loven, L., Russo, W., Di Fatta, G., Riekki, J., and Fortino, G. ”Developing Agent-based Smart Objects for IoT Edge Computing: Mobile Crowdsensing Use Case”, In: IDCS2018, pp. 235-247, Tokyo, Japan, 2018. 9 Thank you for your attention! Questions?