To Cache or Not To Cache: The Effects of Warming Cache in Complex SPARQL Queries

Existing RDF engines have developed caching techniques able to store intermediate results and reuse them in further steps of the query execution process; thus, execution time is speeded up by avoiding repeated computation of the same results. Although these techniques can be beneficial for many real-world queries, the same effects may not be observed in complex queries. Particularly, queries comprised of a large number of graph patterns that require the computation of large sets of intermediate results that cannot be reused, or queries that require complex computations to produce small amounts of data, may require further re-orderings or groupings in order to make an effective usage of the cache. In this paper, we address the problem of determining a type of SPARQL queries that can benefit from caching data during query execution or warming up cache. We report on experimental results that show that complex queries can take advantage of the cache, if they are reordered and grouped according to small-sized star-shaped groups; complex queries are not only comprised of a large number of patterns, but they may also produce a large number of intermediate results. Although the results are preliminary, they clearly show that star-shaped group queries can speed up execution time by up to three orders of magnitude when they are run in warm cache, while original queries may exhibit poor performance in warm cache.