2021/03/23 Tesis Doctoral de U Amozarrain
Fecha de primera publicación: 17/03/2021
Ugaitz Amozarrain Perez: ”Towards a fully mobile publish/subscribe system”.
Zuzendaria_Director: Mikel Larrea Álava.
2021_03_23, 09:00 Informatika Fakultatea. Sala Ada Lovelace Aretoa
Abstract: This PhD thesis makes contributions to support mobility and fault tolerance in a publish/subscribe system. Two protocols are proposed in order to support mobility of all devices in the system, including inside the event notification service. The protocols are designed with the idea that any change due to mobility is completely beyond our control and ability to predict. Moreover, the proposed solutions do not need to know neither the amount of nodes in the system nor their identities before starting, the system is able to adapt to new devices or disconnections and is able to keep operating correctly in a partitioned network. To do so we extend a previously proposed framework called Phoenix that already supported client mobility. Both protocols use a leader election mechanism to create a communication tree in a highly dynamic environment, and use a characteristic of that algorithm to detect topology changes and migrate nodes accordingly. Thus, our first approach was developed with the idea of making the fewest amount of changes to Phoenix while trying to maintain its main benefits. The resulting solution uses a recursive approach to notify migrations and update routing tables where each broker is responsible only for the subscribers that are directly connected to it. The second protocol, that we have decided to call MFT-PubSub, improves on the first one, introducing timestamps and consequently reducing the amount ofmessages to reconfigure the routing tables after a migration. Additionally, the message size is also reduced, and we also prevent possible inconsistencies on partitioned networks. Finally, we prove the validity of MFT-PubSub by simulating it with the Castalia framework for wireless sensor networks and compare it to AODV in terms of performance.