A Reliable Distributed Computing System for Mobile Cloud Computing
Computer Architecture, Embedded and Massively Parallel Systems
Final Report Abstract
First, DiVote was developed, which is a distributed voting protocol in the context of urban polling that is suitable for environments, in which nodes exhibit high mobility. It relies on deviceto-device communication to exchange voting information. The proposed DiVote protocol preserves privacy by using a cryptographic hash function, votes cannot be related to the corresponding node identities. The dynamism due to mobility imposes tight constraints on the convergence speed of the algorithm. Hence, DiVote immediately updates the local estimate. Simulation results obtained when applying DiVote to realistic pedestrian mobility traces show that even in sparse scenarios local estimates quickly converge to the global result after having shared with 30 % of all nodes in the system. At the same time, accuracy of local estimates is ensured as DiVote avoids to count votes multiple times, which is important since the same node may be encountered several times. In dense scenarios, the local estimate does not deviate by more than 3 % from the global result after the system reaches steady state. Rather than storing shared nodes and their votes in plain bit vectors, DiVote uses D-GAP compression to be scalable in terms of required storage capacity. For realistic voting distributions, at least 19 % compression is achieved. Furthermore, the processing overhead introduced at the application layer is very low since only a fraction of the received messages (approximately 30 %) has to be processed even in the densely populated scenarios. Second, we developed UrbanCount, a fully distributed protocol for crowd counting via device-to-device communication, which is suitable for operation in urban environments with high node mobility and churn. Each node collects crowd size estimates from other participants in the system whenever in direct communication range, and immediately integrates these estimates into a local estimate. We evaluated the proposed protocol via extensive trace-driven simulations of synthetic and realistic mobility models, and we answered two main questions: When is distributed counting possible, and how accurate can it be? We showed that for densities above 0.1 nodes/m² distributed crowd counting always produces precise estimates regardless of the area in which mobility occurs. When the density is sufficient, we showed that UrbanCount is able to produce a local estimate with at least 98% accuracy for synthetic and 93% for realistic mobility scenarios. Furthermore, we presented that UrbanCount provides a scalable solution for crowd counting for both indoor and outdoor scenarios. Finally, we showed that the performance of UrbanCount is not affected if the characteristics of mobility in an area are not known in advance. Third, ViPMesh, is a joint emulation/simulation approach with nested virtualization. It relies on WLAN interface emulation and QEMU-based system virtualization with nested container isolation to support early design analysis of real applications on top of an unmodified network stack. This approach allows the evaluation of common ad-hoc, infrastructure-mode, or mesh WLAN setups or any other MAC-layer variant, as configurable under Linux. Adopting an alternative time source approach for QEMU, ViPMesh acts as discrete-event simulator. Furthermore, it integrates comprehensive medium access, channel, and environment models with support for interference effects, IEEE 802.11n MIMO, multi-channel operation, and device mobility. The proof-of-concept implementation of ViPMesh is evaluated in example scenarios, demonstrating its physical layer model. An article reporting about the results of this project has been published on the website of the University of Rostock and in local newspapers. University of Rostock, „Press Releases“: http://www.unirostock.de/detailseite/news-artikel/warnemuender-forscher-entwickelt-neuen-mechanismus/ Ostsee-Zeitung: http://www.amd.e-technik.uni-rostock.de/presse_img/OZ_Tueftler_05102016.pdf
