Replication in loosely coupled systems : simulation and evaluation of replication strategies for their use in loosely coupled IT systems

Modern IT systems are often implemented as loosely coupled systems, consisting of independently developed and deployed components. These systems require a high degree of communication within and can be optimized by the replication of data. Actual research on replication mainly focuses on the data layer, which leads to tight coupling and is therefore not suitable for loosely coupled systems. This thesis identified replication strategies which may be used for loosely coupled systems by a bottom-up approach (based on general characteristics of replication mechanisms) and evaluates their use in respect to qualitative and quantitative indicators. The considered replication strategies mainly build on Event-Driven Architectures with Messaging and Atom-based Web Feeds. Data was gathered by an analytical evaluation and a simulation. Qualitative indicators were solely examined analytically, in accordance with the quality indicators of ISO 25010. Quantitative indicators are the latency of a client-side read operation, the latency of a server-side update operation, the time until all replicas are in a consistent state and the consumed bandwidth. These indicators were examined analytically and measured during a simulation in a custom-built simulation environment with implementation of the replication strategies in Java EE and Ruby on Rails. To emulate the behavior of a real IT system, the simulation uses several synthetic workloads which are based on real-world observations and assumptions made in related research. The evaluation concludes that there is no single best replication strategy for all cases of applications. Replication strategies which use Atom-based Web Feeds are generally not recommended because of the insufficient availability of tools in this area. The evaluation indicates that replication by push-based notifications of the full state of a data item with a publish-subscribe message queue can be used for most cases of applications.