Analysis and optimisation of a genetic algorithm for the generation of specialised replication strategies
Analysis and optimisation of a genetic algorithm for the generation of specialised replication strategies
Background
Data replication is used to minimise the possibility of access failures to urgently required data (high access availability), but also to reduce access times to this data. The System Software and Distributed Systems department recently created a CORBA-based and a JAVA prototype for consistent data replication, which are used to manage replicated WWW documents. These prototypes contain a component that allows any so-called Coterie-based replication process to be specified and used to manage documents on the WWW. The component thus represents a framework for Coterie-based replication procedures. The specification of a desired special replication strategy is carried out by specifying an acyclic graph (in text form according to a formal language).
Furthermore, a first prototype of an automatic designer based on genetic algorithms was developed for such special acyclic graphs. By specifying certain parameters such as the interval of the desired operation availabilities and operation costs (for read and write operations), number of replicas, etc., the automatic designer attempts to find a replication strategy in the form of an acyclic graph that has the desired properties.
Job description
The task to be solved as part of the individual project is the analysis and optimisation of the rudimentary Automatic Designer prototype. It is expected that by analysing the way the prototype has worked so far and its behaviour when creating solutions, conclusions will be drawn regarding necessary improvements. These are then to be integrated into the existing prototype and its improved behaviour documented. In particular, it should be possible to analyse the influence of individual individuals on the development of subsequent populations and whether the selected starting population is "potent" enough to produce offspring with the desired characteristics. Optimisation is intended to compensate for any deficits in the current prototype. The overall improvement achieved in the prototype is to be illustrated and evaluated by before/after synthesis runs with respect to a reference problem set.