A Concurrent and Distributed Analysis Framework for Kieker
Nils Christian Ehmke, Jan Waller, Wilhelm Hasselbring · 2013
Kieker is a framework for monitoring and analyzing software systems. The analyses performed upon the monitoring data are based on pipe-and-filter networks, that are usually executed on a single computer core. In the case of computationally expensive and memory consuming online trace analyses, this basic approach is no longer sufficient. Our approach extends the Kieker framework by adding explicit support for concurrent and distributed analysis networks. The support for concurrency is realized by adding unbounded buffers to the pipe-and-filter networks and by consequently executing the resulting parts asynchronously. Support for distribution is added by aggregating parts of the analysis networks into analysis nodes. These nodes can then be distributed. Our approach is evaluated and benchmarked in a series of thorough lab experiments. These experiments indicate a high communication overhead within our framework modifications. As a result, only little speedup can be achieved in the case of most analyses.