Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 11
Software Engineering Software Engineering Concerns in Observing Concerns in Observing
Autonomic SystemsAutonomic Systems Networks Networks
Prof. A. Taleb-BendiabProf. A. Taleb-BendiabSchool of ComputingSchool of Computing
Liverpool John Moores UniversityLiverpool John Moores Universityemail: email: [email protected]@livjm.ac.ukhttp://www.cms.livjm.ac.uk/taleb/researchhttp://www.cms.livjm.ac.uk/taleb/research
Software Engineering Software Engineering Concerns in Observing Concerns in Observing
Autonomic SystemsAutonomic Systems Networks Networks
Prof. A. Taleb-BendiabProf. A. Taleb-BendiabSchool of ComputingSchool of Computing
Liverpool John Moores UniversityLiverpool John Moores Universityemail: email: [email protected]@livjm.ac.ukhttp://www.cms.livjm.ac.uk/taleb/researchhttp://www.cms.livjm.ac.uk/taleb/research
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 22
OutlineOutline Much is known about model-based autonomic systems design Much is known about model-based autonomic systems design
based on:based on: Feedback control loop, Sensor&Effector, Policy-based managementFeedback control loop, Sensor&Effector, Policy-based management
Centralised and design-time control and monitoring strategies Centralised and design-time control and monitoring strategies for autonomic systems, Etc.for autonomic systems, Etc.
SE techniques exist to handle exhaustive monitoring of a system (e.g. event driven programming, OO design patterns, such as
observer/listener registration), Design by contract, etc. Engineering Networks of Autonomic SystemsEngineering Networks of Autonomic Systems
Are small scale AC systems design models Scalable? Are they applicable to large networks of AC systems
If the exhaustive monitoring of a random network of AC is If the exhaustive monitoring of a random network of AC is computationally expensive then computationally expensive then
whatwhat is the minimal feedback/observation vertices set is the minimal feedback/observation vertices set to enable the monitoring of the whole Networksto enable the monitoring of the whole Networks
Our approach Our approach Algorithm and associated metrics and SE Algorithm and associated metrics and SE support partial support partial
observation of random networks of ACobservation of random networks of AC Experiment and case-studyExperiment and case-study
Conclusions, further works and Q&AConclusions, further works and Q&A
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 33
Large-Scale P2P Large-Scale P2P Self-Self-Organising OverlaysOrganising Overlays
Readings
Show Example-4Monitoring & Sensor & Actuator
Overlay Network
Consumer
Service Provider
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 44
Exploiting Topological Self-Exploiting Topological Self-OrganisationOrganisation
In a Scale-Free Network we can monitor “only” high degree In a Scale-Free Network we can monitor “only” high degree nodes (highly connected) nodes (highly connected) Efficient monitoring of nodes is possible without Efficient monitoring of nodes is possible without
topological knowledgetopological knowledge 17-23% reduction of the number of nodes to be 17-23% reduction of the number of nodes to be
monitored [EASE’07] monitored [EASE’07] ApproachApproach
Random Acquaintance Monitoring methodRandom Acquaintance Monitoring method Characterise the network via SF signatures includingCharacterise the network via SF signatures including
Hub Connection Density metricHub Connection Density metric Acquaintance Nomination metricAcquaintance Nomination metric
Calculate the acquaintance nomination graphCalculate the acquaintance nomination graph Adapted acquaintance immunisation algorithm [Adapted acquaintance immunisation algorithm [R Cohen et
al. 2002]] Selection nominated nodes (often hubs) Selection nominated nodes (often hubs)
Establish observer graph (overlay) Establish observer graph (overlay) Deploy typed observer agents.Deploy typed observer agents.
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 55
So Far …So Far … Simulation framework Simulation framework Network CharacterisationNetwork Characterisation
Such as acquaintance monitoring Such as acquaintance monitoring algorithm, metrics, etc.algorithm, metrics, etc.
Meta-language for network Meta-language for network governance and observationgovernance and observation A method of externalising observation A method of externalising observation
and deployment logic for ease of self-and deployment logic for ease of self-adaptation and evolutionadaptation and evolution
Software design patterns for observer Software design patterns for observer agent generation and deploymentagent generation and deployment
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 66
Case-Studies: P2P Network Case-Studies: P2P Network SimulationSimulation
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 77
Case-Studies: Salt World Case-Studies: Salt World TestbedTestbed
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 88
ConclusionsConclusions We argued that AC design model can be used to We argued that AC design model can be used to
design large-scale networks of AC systems design large-scale networks of AC systems When we used Random Acquaintance Monitoring When we used Random Acquaintance Monitoring
methodmethod As monitoring/feedback of scale-free autonomic networks As monitoring/feedback of scale-free autonomic networks
can be improved by reducing the number of monitored nodes can be improved by reducing the number of monitored nodes with near complete network knowledgewith near complete network knowledge
Further work is requiredFurther work is required To evaluate the method in other case-studies To evaluate the method in other case-studies
includingincluding The impact of heterogeneity of monitored AC nodes The impact of heterogeneity of monitored AC nodes reasoning issues. reasoning issues.
Extension of the method to other network types Extension of the method to other network types including random, regular, etc. including random, regular, etc.
Enhance the software framework.Enhance the software framework.
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 99
That’s the end – so I’m off !
Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: Prof. A. Taleb-Bendiab, Talk: SOAS’07, Contact: www.livjm.ac.uk/taleb/research, Date: 21/04/2321/04/23, Slide: , Slide: 1010
AcknowledgementsAcknowledgements Acknowledgements
My thanks to the Team
Useful Links www.cms.livjm.ac.uk/2nrich www.cms.livjm.ac.uk/cloud www.cms.livjm.ac.uk/taleb