Measuring the relative impact of network overhead and client-side query optimisation in link traversal over Linked Data documents

Data decentralisation initiatives, while addressing issues around regulatory compliance and barriers to entry into data-driven markets, also introduce new technical challenges with regard to data access. These challenges are imposed upon the data access abstraction layer, such as client-side query engines, that become responsible for performant data access. Especially in the context of interactive applications, user-perceived sluggishness can inhibit the adoption of applications built on decentralised data, and, by extension, the decentralisation initiatives themselves, should they appear infeasible for such use cases. The performance of the data access layer is the combination of data retrieval over the network and the processing of this data locally. Prior work has demonstrated how the structure of certain decentralised environments can assist query engines in efficiently locating and accessing query-relevant data, reducing the relative impact of data access, and exposing the local processing as a major bottleneck. Within this work, we measure the impact of network overhead on query execution over decentralised Linked Data documents, when combined with an example restart-based adaptive query planning approach, using the Solid ecosystem as an example environment. Through the addition of rate-limiting and network latency simulation, we are able to demonstrate how client-side query optimisations that provided considerable improvements under optimal network conditions in prior work are unable to produce measurable improvements under more realistic conditions. Our work illustrates the importance of using realistic benchmarking environments, and highlights the importance of future work on reducing or circumventing the network overhead as a whole in link traversal.