Consistency and Fault Tolerance of Distributed Storage Systems

Application of erasure coding for fault tolerance in distributed storage systems.

Distributed data storage systems comprise a large number of network-connected resources to provide large and extendable data storage capacities. The system becomes more vulnerable to faults in any of its components as it increases in size. This results in an increasing failure probability and risk of data loss. A commonly and widely applied fault tolerance technique is replication. A viable alternative is to use coding techniques, like Maximum Distance Separable (MDS) erasure codes that provide reliability with an optimal redundancy ratio. In this thesis we analyse the application of erasure coding as a technique to provide redundancy for fault tolerance in distributed storage systems. The focus is on the influence of erasure coding with varied redundancy ratios on the data storage reliability, on data access protocols that guarantee consistency with concurrent updates and failing clients or storage servers and on recovery cost in case of component faults.

Autorentext
Kathrin Sobe studierte Informatik an der Universität zu Lübeck. Von 2005 bis 2011 war sie als wissenschaftliche Mitarbeiterin am Zuse Institut Berlin in mehreren Forschungsprojekten im Bereich Grid Computing und verteilte Datenspeichersysteme tätig. Von 2011 bis 2012 arbeitete sie an der Humboldt Universität zu Berlin im Rahmen eines DFG Graduiertenkollegs und stellte ihre Dissertation fertig. Seit 2013 ist sie als Softwareentwicklerin tätig und lebt in Dresden.