Current research in the area of self-adaptive systems is moving towards solutions to adaptation problems with the aim to engineer systems that can quickly respond to changes without any human intervention. As systems are becoming larger and more complex, the adoption of solutions that are both decentralised and scalable is becoming increasingly important. Up to now, this research area has focused on producing approaches to support the actuation of decentralised adaptation actions, however the problem of deciding when and how to execute them is still challenging in a decentralised setting. SPANDO proposes to solve such problem by using performance prediction models that are being studied in operations research and applied probability research. The most common prediction models that are already being used at run-time are based on Continuous-Time Markov Chains. However, these existing techniques have scalability limitations due to the state-space explosion of the CTMC formalism.
The SPANDO project will overcome these limitations by studying a new class of performance prediction models that can be evaluated in a decentralised way, without any explicit coordination. The proposed models will use formalisms based on ordinary differential equations, such as fluid and mean-field analysis, and have the particularity of being independent of the size of the system. The results of the evaluations of these models will then be used at run-time as inputs for proper decentralised adaptation actions.
Marie Curie Fellow: Dr. Daniel J. Dubois
Scientist in Charge: Dr. Giuliano Casale
Institution: Imperial College London, Department of Computing
Start date: 01/12/2014
End date: 30/11/2016