|Title:||System Administration and Emergent Properties
of Pervasive Computing Environments
|Funding:||Oslo University College/The Ministry
of Education and Research
PhD project description
In a pervasive computing scenario, cooperative behaviour cannot be taken for granted. Devices are controlled by many autonomous users, bound by no centralized authority. To be able to predict various aspects of agent behaviour, some kind of tool is needed, in order to be able to model agent environments, both constraints and interactions, as well as policy. Moreover, we need to be able to analyse these models, to be able to predict the outcome of different ’initial configurations’. Promise theory is a new theory about what can happen in a network of entirely autonomous components. Rather than adopting the conventional belief that “only that which is programmed happens”, it takes the opposite viewpoint: “only that which is promised can be predicted”. It therefore approaches management from the viewpoint of uncertainty with realism rather than faith in compliance. Promise theory is a graph-theoretical framework that can be used for modelling (interacting) autonomous entities, where each promise represents a potential transfer of value between agents. This transfer might imply a reduction or increase in the possessed resources of each agent, dependent on whether the agent provides or receives a service as a result of the promise. The idea is to model the potential risks and benefits present in an environment of autonomous entities, which typically is the source of cooperative or non-cooperative agent behaviour. The promises then can be viewed as a kind of ’measuring stick’ for cooperative behaviour, as the established promises clearly defines what is assumed cooperative. We wish to show that promise theory helps one to understand the cooperative properties of autonomous agents for the purpose of determining optimal policies for their management. Here there are some similarities with the multi-agent concept of commitments. A large part of this research project regards developing this framework further.
- Relationship between games and promises. We wish to take advantage of game modelling in order to predict what kind of agent behaviour is beneficial for each of the entities. To accomplish this fully, we need to express our promise networks in a game-theoretical context. The goal then is to evaluate the outcome of different promise graphs, and predict probable behaviour based on these.
- Relationship between policy and promises. As we have based our work on the policy-based management paradigm, we need a consistent way of expressing agent and system policy in terms of promise theory.
- Modelling interaction. Agent interaction involves a number of important parameters: time and iterations, transfer of value (common currency), trust, etc. How can we calculate the probability of an agent cooperating?
- Burgess, M., Fagernes, S. ’Autonomic Pervasive Computing: A Smart Mall Scenario Using Promise Theory’. Modelling Autonomic Communications Environments (MACE06), Dublin, Ireland, October 25-26. (refereed proceedings)
- Burgess, M., Fagernes, S. ’Promise theory – a model of autonomous objects for pervasive computing and swarms’. World Class Technology Summit (WCTS06), Silicon Valley, USA, July 16-23. (refereed proceedings)
- Begnum, K., Burgess, M., Jonassen, T., Fagernes, S. ’On the Stability of Adaptive Service Level Agreements’. (IEEE Transactions on Network and Service Management)
- Begnum, K., Burgess, M., Jonassen, T., Fagernes, S. ’Summary of Stability of Adaptive Service Level Agreement’. Sixth IEEE International Workshop on Policies for Distributed Systems and Networks (POLICY05), Stockholm, Sweden, June 6 - 8, 2005 (refeered proceedings).
- Burgess, M. Fagernes, S. ’The effects of tit for tat policy for rejecting spam or denial of service floods ’. System Administration and Network Engineering (SANE2004), Amsterdam, Netherlands, Sep. 27 - Oct 1, 2004 (refereed proceedings).
- [_URL_ Homepage] of Siri Fagerness
- Publications of Siri Fagerness, as indexed by DBLP