Enhanced adaptive partitioning in a distributed graph database

What is it about?

Nowadays, open-source graph databases do not include an inherent mechanism for data relocation that would be based on their usage. They often do not offer even appropriate monitoring that could help to make such a decision. Information about data utilization could, however, work as an input to some decision-making process about more suitable data regrouping that could be much more efficient in terms of intra-network communication. Therefore, we created a module for the graph computational framework TinkerPop that logs traffic generated by the user queries. These logged records serve as an input for the algorithm of Adaptive Partitioning that we enhanced with better balancing, avoidance of local optima and the notion of weighted graphs. This approach yields a 70–80% improvement in intra-network communication, which is comparable to other methods, namely Ja-be-Ja, that offers similar results but has higher computational demands.

Featured Image