Adaptive Data Distribution for Cost- and Workload-aware Data Management in the Cloud (PhD Thesis, ongoing)

In this thesis, we will extend the ClouDMan cost model which considers data replication or data partitioning (exclusive or) by combining both. Especially for mixed workloads, it is beneficial to combine partial replication and data partitioning, to provide a high degree of availability (replication) without the need for synchronizing changes in the entire system.

The first version of the cost model that will be developed will be a static one. In the second version, support for workload monitoring and dynamically changing workloads will be added. Additional aspects that will be considered in the second version include support for different versions per object (i.e., data freshness), which will be another parameter for users to specify their data requirements.

 

In summary, this thesis has the following goals:

  1. Develop a first version of the cost model that allows to recommend a hybrid data replication/data partitioning schema for static workloads.
  2. Develop a dynamic version of the cost model that is able to cope with dynamic workload changes and that is able to recommend changes in the data distribution.
  3. Add support for data freshness to the model, allowing additional types of data accesses (up-to-date data vs. data with lower freshness levels).

Staff

Research Topics