using trust-aware strategic agents for a self-organising computing grid
DESCRIPTION
Presentation by Yvonne Bernard at the 2nd Awareness Workshop on Challenges for Achieving Self-awareness in Autonomic Systems @ SASO 2012, Lyon, FranceTRANSCRIPT
Y. Bernard 10.09.2012 Awareness PhD Forum
1
Using trust-aware strategic agents for a self-organising computing grid
2
Y. Bernard An Evolutionary Approach to Grid Computing Agents 2
Y. Bernard 10.09.2012 Awareness PhD Forum
Outline
q Motivation: Organic Compting system class
q Trust in OC systems
q Application scenario: Trusted Desktop Grid
q Contribution
q Agent types and hierarchy [10]
§ Static agents [3]
§ Trust-adaptive Agents
- iTC Agents [5],[7]
- Evolutionary Agents [9] § Strategic Agents [8]
q Adaptive Model of Observation
q Summary and Outlook
3
Y. Bernard An Evolutionary Approach to Grid Computing Agents 3
Y. Bernard 10.09.2012 Awareness PhD Forum
q New way of dealing with complexity
§ Self-X properties for decentralised solutions
§ Incomplete system information
§ Manage opennes - Autonomous unknown agents - Selfish agents - Malicious agents
q Implications to Trust facettes:
§ Reliability/Functionality: Dynamic structure requires new approaches § Security: Privacy and cooperation at the same time
§ Safety: corrections during runtime possible
§ Credibility: analyse environment at runtime
§ Usability: Transparency and predictability not guaranteed
à New class of algorithms necessary
Agent B
Motivation: Organic Computing system class
Agent C
XAgent A
Agent D
Example: Open Desktop Grid
4
Y. Bernard An Evolutionary Approach to Grid Computing Agents 4
Y. Bernard 10.09.2012 Awareness PhD Forum
Trust
q Trust := expectation value § Probability that a certain event will happen in the future
§ Reputation := Trust from indirect experience
q Trust is a social mechanism, which allows more efficient and effective cooperation between individuals.
q This mechanism can be transferred into technical systems.
§ Include trust aspect in cooperation decision
q Trust as a constitutional part of technical systems
§ Reduces information uncertainty in open systems (e.g. OC)
§ Enables cooperation between subsystems (agents)
- increase efficiency of cooperating agents
- increase robustness regarding misbehaving agents
5
Y. Bernard An Evolutionary Approach to Grid Computing Agents 5
Y. Bernard 10.09.2012 Awareness PhD Forum
Application Open Desktop Grid Computing
§ Computation on computers from different domains:Open system § Free-riders refuse to accept work units. § Egoists return wrong/incomplete results.
§ Requires job replication and result checking -> inefficient
Agent B
Agent A
Agent C
Agent D Agent E
Agent F
Agent G
F E
6
Y. Bernard An Evolutionary Approach to Grid Computing Agents 6
Y. Bernard 10.09.2012 Awareness PhD Forum
Trusted Desktop Grid
q Decentralized system: All agents can
§ offer computing resources (worker) and/or
§ submit work units (submitter).
q Autonomous agents act on behalf of the users.
q Agents have a motivation to cheat.
q Basic Idea: enhance matchmaking with trust information § Submitter: Who will be asked to process work units?
§ Worker: Whose work units to accept?
q Goal: Enhance efficiency and robustness using trust and adaptation.
7
Y. Bernard An Evolutionary Approach to Grid Computing Agents 7
Y. Bernard 10.09.2012 Awareness PhD Forum
Contribution
q Architecture for trust-adaptive strategic agents
§ Generic model for OC and adaptive systems
- Adaptive to current situation
- Strategic decision making
- Model of Observation
- Institutional contol using constraints
q Implementation of local trust-based adaptive strategy algorithms
§ Submitter strategies § Worker strategies
q Evaluate architecture and matchmaking strategies and compare to Related Work (based on Grid metrics)
§ H-Trust[12]: Trust- and Credibility-Tables
§ Organic Grid[11]: Adaptive Tree Overlay
8
Y. Bernard An Evolutionary Approach to Grid Computing Agents 8
Y. Bernard 10.09.2012 Awareness PhD Forum
Agent types P
erfo
rman
ce
Awareness
Workload +Trust/Rep. +SD.S
High throughput/time Short makespan Decreased waste Decreased replication overhead
+SD.L
à Increasing observation overhead!
9
Y. Bernard An Evolutionary Approach to Grid Computing Agents 9
Y. Bernard 10.09.2012 Awareness PhD Forum
Agent types P
erfo
rman
ce
Awareness
Workload +Trust/Rep. +SD.S +SD.L
Trust-neglecting
agent
Trust-aware agents
Reactive trust-adaptive agents
iTC agent Evolutionary agent
Fixed stereotype agents
Pro-active trust-strategic agents
Tactical agent Adaptive MoO agent eTC agent
10
Y. Bernard An Evolutionary Approach to Grid Computing Agents 10
Y. Bernard 10.09.2012 Awareness PhD Forum
Agent hierarchy
q Static trust-considering agents: § Behaviour prototypes:
Free Rider, Egoist
q Trust-adaptive agents: reactive
§ Adapt parameters to situation
§ iTC Agent
§ Evolutionary Agent q Trust-strategic agents: proactive
§ Tactical Agent: includes other agents‘ expected behaviour
§ eTC Agent: includes institutional control
§ MoO Agent: Long-term strategic behaviour (access to predictions)
- Aim: find suited information/solution quality relation regarding overhead à adaptive Model of Observation
11
Y. Bernard An Evolutionary Approach to Grid Computing Agents 11
Y. Bernard 10.09.2012 Awareness PhD Forum
Adaptive Model of Observation
q Only evaluate information necessary for the current situation
q Overall goal: Reduce Overhead without sacrificing solution quality
q Types of Overhead
§ Communication:
- Update frequency (e.g. of reputation values)
- How many agents are asked to determine certain values (e.g. workload)?
§ Calculation/storage:
- Aggregation of values
- Storing values for further evaluation (e.g. Time series analysis for prediction, relevant for strategic level)
q Adaptive cognition: select observed parameters based on
§ role (submitter, worker) and
§ situation (normal, increased attentiveness, alert)
12
Y. Bernard An Evolutionary Approach to Grid Computing Agents 12
Y. Bernard 10.09.2012 Awareness PhD Forum
q Trust can enhance communication, collaboration and negotiation in complex systems (e.g. OC systems)
q Application scenario Trusted Desktop Grid
q Approaches to trust-adaptive strategic agents [10]
§ Static agents[3]: stereotypes of agent behaviour § Trust-adaptive agents
- iTC Agents [5],[7]
§ Efficient and robust
§ Planned: Optimisation using learning techniques (thresholds)
- Evolutionary Agents [9]: first distributed learning approach
§ Strategic Agents
- First approach: Tactical agent[8]
- Planned: eTC and MoO agent:
§ Strategic Level on top of iTC Agents, institutional constraints
§ Strategy based on long-term data and prediction
Summary and Outlook
13
Y. Bernard An Evolutionary Approach to Grid Computing Agents 13
Y. Bernard 10.09.2012 Awareness PhD Forum
Thank you for your attention!
14
Y. Bernard An Evolutionary Approach to Grid Computing Agents 14
Y. Bernard 10.09.2012 Awareness PhD Forum
Publications q [1] Martin Hoffmann, Michael Wittke, Yvonne Bernard, Ramin Soleymani, Jörg Hähner, “DMCtrac:
Distributed Multi Camera Tracking,”
ICDSC ’08. Second ACM/IEEE International Conference on
Distributed Smart Cameras, Sept. 2008.
q [2] Sven Tomforde, Martin Hoffmann, Yvonne Bernard, Lukas Klejnowski and Jörg Hähner, "POWEA: A System for Automated Network Protocol Parameter Optimisation Using Evolutionary Algorithms",
Beiträge der 39. Jahrestagung der Gesellschaft für Informatik e.V. (GI), 2009,
pp. 3177--3192, Gesellschaft für Informatik e.V. (GI)
q [3] Yvonne Bernard, Lukas Klejnowski, Jörg Hähner, Christian Müller-Schloer, "Towards Trust in Desktop Grid Systems", ccgrid, pp.637-642, 2010 10th IEEE/ACM International Conference on Cluster, Cloud and Grid Computing, 2010
q [4] Jan-Philipp Steghöfer, Rolf Kiefhaber, Karin Leichtenstern, Yvonne Bernard, Lukas Klejnowski, Wolfgang Reif, Theo Ungerer, Elisabeth André, Jörg Hähner, and Christian Müller-Schloer, "Trustworthy Organic Computing Systems: Challenges and Perspectives", Proceedings of the 7th International Conference on Autonomic and Trusted Computing (ATC 2010), Springer
q [5] Lukas Klejnowski, Yvonne Bernard, Jörg Hähner and Christian Müller-Schloer, "An architecture for trust-adaptive agents", Proceedings of the 2010 Fourth IEEE International Conference on Self-Adaptive and Self-Organizing Systems Workshop (SASOW 2010)
15
Y. Bernard An Evolutionary Approach to Grid Computing Agents 15
Y. Bernard 10.09.2012 Awareness PhD Forum
Publications q [6] Jan-Philipp Steghöfer, Florian Nafz, Wolfgang Reif, Yvonne Bernard, Lukas Klejnowski, Jörg
Hähner and Christian Müller-Schloer, "Formal Analysis of Trusted Communities", Proceedings of the 2010 Fourth IEEE International Conference on Self-Adaptive and Self-Organizing Systems Workshop (SASOW 2010)
q [7] Yvonne Bernard, Lukas Klejnowski, Emre Cakar, Jörg Hähner and Christian Müller-Schloer, "Efficiency and robustness using Trusted Communities in a Trusted Desktop Grid", Proceedings of the 2011 Fifth IEEE International Conference on Self-Adaptive and Self-Organizing Systems Workshop (SASOW 2011)
q [8] Yvonne Bernard, Lukas Klejnowski, Ronald Becher, Markus Thimm, Jörg Hähner, Christian Müller-Schloer, "Grid agent cooperation strategies inspired by Game Theory", 4. Workshop Grid-Technologie für den Entwurf technischer Systeme, Dresden, 21.-22. September 2011, ISSN 1862-622X
q [9] Yvonne Bernard, Lukas Klejnowski, David Bluhm, Jörg Hähner and Christian Müller-Schloer, "An Evolutionary Approach to Grid Computing Agents", Proc. of the Italian Workshop on Artificial Life and Evolutionary Computation, 2012 , pp. 1-12, ISBN 978-88-903581-2-8
q [10] Yvonne Bernard, Lukas Klejnowski, Jörg Hähner, and Christian Müller-Schloer, "Self-organising Trusted Communities of Trust-adaptive Agents", Awareness Magazine 2012, www.awareness-mag.eu, doi: 10.2417/3201011.004065
q [11] A.J. Chakravarti, G. Baumgartner and M. Lauria. „The organic grid: self-organizing computation on a peer-to-peer network“. In: IEEE Transactions on Systems, Man and Cybernetics, Part A: Systems and Humans, 35.3 (Mai 2005), S. 373 –384. issn: 1083-4427. doi: 10.1109/TSMCA.2005.846396.
q [12] Huanyu Zhao and Xiaolin Li. „H-Trust: A Robust and Lightweight Group Reputation System for Peer-to-Peer Desktop Grid“. In: 28th International Conference on Distributed Computing Systems Workshops. ICDCS ’08. Juni 2008, S. 235 –240.
16
Y. Bernard An Evolutionary Approach to Grid Computing Agents 16
Y. Bernard 10.09.2012 Awareness PhD Forum
Outlook
q Controller
§ Parameter optimisation (learning) on operational level
§ Add long-term strategies (strategy)
- influence operational level
§ Institutional control: Constraints
- Pre-filtering
- Post-filtering
q Observer
§ Adaptive Model of Observation regarding: - Which parameters are observed?
- Update frequency
- Agents sample size
- Memory size
- Aggregation method (time series analysis, Neural Networks,…)
q Compare trust-strategic agent with related work (H-Trust, Organic Grid)
17
Y. Bernard An Evolutionary Approach to Grid Computing Agents 17
Y. Bernard 10.09.2012 Awareness PhD Forum
Trusted Manager O C
Agent
Operational level Observer Current Situation
Controller
Operational Decision S
W
Productive level Observer Internal Situation
Controller Productive Interaction
Submitter
Worker
Strategic level
Observer Long-term Situation
Controller
Strategic Decision S W
SD.S WLTC,
TrustAgents, RepAgents, RepOwn, Fitness
SD.L Predict(WL), Predict(Trust) Predict(Rep)
Pre-selected Behaviour
Behaviour
Constraints