Martin Junghanns

Currently we implement `WritableComparable`. We should benchmark how this behaves when it onlyimplements Flink's `Value` interface (and of course `Comparable`). For GradoopId, it turned out that implementing only `Value` leads...

The current matching strategies are isomorphism and homomorphism which refer to both, vertex mappings and edge mappings. However, homomorphism and isomorphism should also be configurable individually for vertices and edges.

Flink is able to run multiple jobs sequentially from a single benchmark class, thus a dedicated and hard-to-maintain bash script is not necessary.

The internal API mainly contains the operator interfaces, the graph representations and the call methods. The current convenience methods at `LogicalGraph` and `GraphCollection` reflect the DSL visible to the user....

Currently, `GraphCollection.distinct()` compares logical graph based on their graph id. The comparison needs to be modularized. distinctBy... - graph head id - element ids - vertex ids - isomorphism

The current implementation creates a new edge data set in each iteration (because of edge directions). One naive approach would be to duplicate (and reverse) edges before the iteration starts...

We need a concept for storing lineage information for logical graphs. This is useful to track from which logical graph a logical graph has been created. One way could be...

Benchmark datasets

3

I was wondering if you could share the benchmark datasets from the paper and the queries you ran for the evaluation. Alternatively, you might wanna share the script that converts...