Anticipating unWanted behavior in Autonomic Computing Systems

Start date: 01.01.2007
End date: 31.12.2010
Duration: 24 months
Funded by: German Academic Exchange Service
Local head of project: Dipl.-Inf. Holger Kasinger
Local scientists: Prof. Dr. Bernhard Bauer
External scientists / cooperations: Prof. Dr. Joerg Denzinger (University of Calgary)
Publications: Publication list



From a developer unintended and not predictable states can cause self-organizing, emergent and thus autonomic systems at runtime to exhibit an unwanted and harmful behavior due to wrong self-adaption. The objective of the AWACS project is to proactively determine such critical system states at runtime before they occur as well as to avoid the unwanted behavior by means of appropriate self-adaption mechanisms.


Unexpected and not predictable environment situations at runtime may trigger even a correctly working advisor to adapt a self-organizing emergent system into previously unknown system states. On the one hand side unknown states are desired or rather required in order to adapt the system to dynamically changing or unexpected situations. On the other side unknown states are undesired if they cause a system to show an unwanted behavior, e.g. inefficiencies in problem solving or even system crashes. However, prohibiting unknown states or returning from them too early limits the adaptivity of the system, whereas staying too long may by contrast render such a returning void. Thus, the objective of the AWACS² (Anticipating Unwanted behavior in Autonomic Computing Systems) project is, based on the results of the AWACS project, to anticipate unwanted behavior at runtime way ahead of its occurrence and to give an advisor enough time to prevent or avoid misleading adaptations autonomously, which promotes the trustworthiness of advised self-organizing emergent systems. We therefore develop a test system embedded in the advisor that continuously tests possibly awkward environment situations, which will lead to unknown system states respectively unwanted behavior, in advance and, if applicable, triggers the advisor to adapt its advising accordingly.