|
The goal of this thesis work is to find a distributed solution for
the propagation manager of SIES, to improve the throughput and
availability of the whole system.
SIES is the abbreviation for Stuttgart Information and Exploration
System. It is an instance of Enterprise Application Integration
(EAI) in data level in the context of heterogeneous industrial IT
infrastructures. Based on the concept of data change propagation,
SIES propagates the data changes from source systems to dependent
target systems, according to the predefined dependencies. This
approach has highly saved the development expenses comparing with
other EAI approaches. However, as a single-node solution, the
throughput of the system is still limited. In previous works, one of
the authors demonstrated that the complex propagation process, in
which many XML objects processing are involved, is the main reason
of low performance. Optimization on the program reduced the average
processing time, but did not help a lot to improve the system
throughput. To meet the requirements from the industry, an extended,
scalable and reliable solution is therefore required for future.
This thesis explores possible parallel/distributed approaches for
extending SIES, especially Propagation Manager. After comparing
alternative solutions, a central load manager based solution is
selected, in which congenetic propagation, m-to-n propagation and
load balancing are well managed. Based on this, the concept of
Propagation Service Unit (PSU) is proposed as a scalable solution
for further extension. Evaluation shows that load dispatching
algorithms in Load Manager are efficient.
|