DISS . ETH NO . 16805 Distributed Coordination Collecting , Locking , and Matching

Distributed computing systems consist of numerous individual computing devices (e.g. computers or sensor nodes) which can communicate with one another. A listing of prominent and exemplary distributed systems would surely have to include rather tightly-coupled shared memory multiprocessor machines and loosely-coupled systems like the Internet or peer-to-peer systems. In general, each entity in a distributed system has its own tasks, which are usually independent of the other entities’ tasks. May the tasks be uncorrelated, this in no case implies that also the actions of the entities are independent. In contrary, for various reasons, such as resource sharing, availability, and fault-tolerance, these entities have to coordinate their actions. This dissertation studies three of the many aspects of the coordination problem in distributed systems: collecting, locking, and matching. In the collect problem processes need to announce decision values to all other processes and conversely collect decision values from all other processes. The values are written in registers located in a shared memory region. This dissertation proposes a range of algorithms for the problem, each featuring novel approaches, including randomization. The second problem, locking, is a generalization of dining philosophers: Each participating process needs to lock multiple registers located in a shared memory region in order to perform some operation on the locked registers. The problem is on the one hand studied theoretically, on the other hand it is also examined from a practical perspective by implementing and evaluating the suggested algorithm. In the last part of the dissertation several matching variants are considered. In particular, distributed matching, minimum weighted perfect matching, and fair bipartite matching are introduced and analyzed.