KubeGraphBench: Benchmarking graph databases for Kubernetes observability

Graph Databases (Graph DBs) are increasingly adopted due to their efficiency in handling complex, interconnected data. They offer a robust alternative to traditional relational Databases (DBs), which often struggle with evolving schemas and deep, multi-hop relationships. The objective of this work is to evaluate the performance and capabilities of two different Graph DB technologies, namely Neo4j (Neo4j) [1] and Memgraph (Memgraph) [2]. The performance evaluation presented builds upon existing infrastructure for constructing and maintaining a knowledge graph representation of Kubernetes clusters. We focus solely on benchmarking the read and write performance of the DBs under increasingly larger data workloads, capturing metrics including latency, memory and CPU usage, and throughput. The DBs are tested fairly while ensuring they are placed under the same set of conditions. The differences in technologies between the DBs are understood and have been taken into account before drawing conclusions. Only the freely available versions of the DB technologies are used. The drawbacks of these versions are taken into account compared to their paid counterparts. Through experimentation, it is concluded that Memgraph, the in-memory storage graph database, outperformed Neo4j in write-heavy benchmarks, but during read-heavy benchmarking, it only outperforms Neo4j in the case of small graphs or simple queries.