Dynamic Freshness-Aware Replica Management and Archiving in Data Grid and Data Cloud Environments (Master Thesis, Finished)

Author

Filip-Martin Brinkmann

Description

The main question the thesis seeks to answer is how one can achieve reliable archiving of data in the cloud. The main challenges arise from the fact that when dealing with relaxed consistency, multiple versions of the same data concurrently exist in the cloud. We are convinced that one may exploit this property in order to make archiving not only feasible, but also efficient. Certainly, a large number of possibilities to achieve this goal exist. Therefore, the thesis shall evaluate different approaches and implementations. As the project builds on Re:GRIDiT, it has to deal with properties specific to Re:GRIDiT. However, we hope to find generic mechanisms and properties that are not restricted to Re:GRIDiT, but may be applied to other data grid / cloud architectures as well. Another use-case we identified is the need for conducting timeline-analyses or trends of data items. By using the multi-version property of the system, it may be possible to implement a minimal-invasive way to fetch recent versions of the same data item. Since already without an archiving system, the system holds different versions of data, it should be possible to collect those versions and reconstruct the data’s timeline. Depending on how old the earliest version being queried is, this timeline may be aggregated over logical logs, concrete archived versions or a mixture of those sources. As it is certainly necessary to restrict the system in order to control which data needs to be archived and which does not, some control mechanism must be implemented. Furthermore, this system may then be used to express requirements on the data and to apply advanced policies to them. This novel possibility may even lead to such a finegrained control over the archiving process that it becomes possible to compute the costs for archiving data. Download Thesis: MSc_Thesis_Brinkmann

Start / End Dates

2010/11/29 - 2011/05/28

Supervisors

Research Topics