A new algorithm for small-large table outer joins in parallel DBMS

Robust and Efficient Large-Large Table Outer Joins on Distributed Infrastructures

“…Moreover, compared with the state-of-art techniques PRPD+DER [4,15] described in Section 2, our approach does not involve any redundancy in join (or lookup) operations, because our method is totally duplication-free and all nodes only receive the tuples that they will eventually use. This should make the approach more efficient, and we will conduct a detailed performance comparison in Section 5.…”

“…State-of-the-art methods designed specific for outer join implementation achieve significant performance improvements [4], however, they are based on the duplication-based method and cannot be applied to large-large table outer joins.…”

“…Often, the assumption is that inner join techniques can be simply applied to outer joins, as identified in [4]. However, applying such techniques for outer joins directly may lead to poor performance [16].…”

“…DER. Xu et al [4] also propose another algorithm called DER (duplication and efficient redistribution), aimed at optimizing outer joins. This method comprises two stages: (1) R i at each node i is duplicated to all the nodes to start inner joins.…”

“…For example, for a left outer join ( ) between two inputs R and S on their attributes a and b, the following query returns not only the matched tuples in the form of <x,a,y>, but also <x,a,null>, when values do not match. select R.x R.a S.y from R left outer join S on R.a = S.b (Query 1) Currently, as for inner joins, implementations for distributed outer joins utilise one of two distributed patterns [4]: redistribution-based and duplication-based outer joins. To study the core performance characteristics of these approaches, we focus on analyzing the parallelism within a single outer join operation between two relations R and S on an n-node system (assuming both R and S are in the form of <key, value> pairs and |R| < |S| in the following).…”