This repository contains the implementation of algorithms for the construction of DecoPa plans, tools for the generation of input streams and a Flink-based implementation for parallel CEP matching.
The directory DecoPa contains the implementation of our algorithms and some of the scripts used to conduct the experiments presented in the paper.
Moreover, the subdirectory DecoPa/poisson-event-gen contains the tools to generate local input streams for a set of processing units as they would be provided by a shuffler, based on a given DecoPa plan.
To this end, either a poisson process is envoked for an event rate distribution or, the path to a .csv-file comprising a global input stream is passed. The latter was done for the two real-world data sets citibike and google cluster, which are also contained in the directory (task\_events\_48h\_2023\_06\_10\_.tar.xz and citibike\_1d1min\_2023\_10\_10.tar.xz).
The script ./findScaling_dop_binary_single.sh can be used, to generate DecoPa plans (and respective local input streams) for a query of length 6 and for clusters with a increasing number of processing units. After the termination of the script, the resulting plans with respective scaling factors can be found in DecoPa/plans/autoscaling.
The directory CEP_Flink_Engine contains the implementation of a Flink-based implementation for parallel CEP matching. The directory CEP\_Flink\_Engine/deploying/example\_inputs contains the DecoPa plans and local streams used for the real-world data experiments. The directory deploying contains the scripts to deploy a disseminated DecoPa plan on a set of processing units.