Mobile Ad-Hoc Networks (MANETs) are self-organized wireless networks where mobility and limited energy resources cause frequent communication and node failures. Guaranteeing consistency and integrity of distributed data in such a volatile environment is challenging. A key concept to assure these guarantees are distributed atomic transactions. Transferring this concept to a MANET environment raises several new research questions due to high failure probabilities. This work analyzes blocking risks of distributed transactions in MANETs and provides solutions to control these risks.


It is well known that a non-blocking atomic commit protocol cannot exist in presence of communication and node failures. This impossibility has little impact on transaction processing in fixed networks, since communication and node failures are so rare that transaction processing is not significantly affected or delayed; however, the situation in MANETs is not clear. Research has not answered yet which transactions show high blocking risks and how blocking risks are influenced by different transaction models. Therefore, a controlled risk management in MANET transaction processing is not possible yet. This thesis contributes towards a better understanding of atomic transaction processing in MANETs by presenting:

• A probabilistic model to predict the abort and blocking risks for arbitrary transaction and MANET scenarios caused by communication or node failures. The model is used to analyze strict and semantic transaction models.




• A solution to control blocking risks caused by participant failures called Shared Log Space (SLS). The SLS system allows to preserve decision logs of a transaction at a defined availability within a MANET for recovering participants. It is shown how the SLS is embedded in commit processing of strict and semantic transactions and how blocking risks can be decreased to a desired level. Two implementation approaches of the SLS are described and evaluated.




• A probabilistic model to analyze the use of a backup coordinator (BC) to reduce blocking risks caused by a node failure of the transaction coordinator. It is shown that an integration of a BC is not necessarily beneficial but may increase the blocking risk in some situations. The presented probabilistic model allows to identify such situations. Additionally, I propose a scheme to integrate the BC with the SLS to assure a required availability of decision logs if the transaction coordinator fails.