Correlated Query Process and P2P Execution

Abstract

A technical trend in supporting data intensive applications is to push-down them to database engines by wrapping computations with User Defined Functions (UDFs) callable from the SQL interface. For seamlessly integrating applications into SQL queries, we introduce a special kind of UDFs with both input and output as relations, or row sets, called Relation Valued Functions (RVFs). As a single SQL statement has limited expressive power on complex data flows and control flows, we propose the notion of Correlated Query Process (CQP), for specifying an application at the process level, by several queries and RVFs with interleaved steps.

We further extend this notion to Collaborative CQP with multiple participating parties which have correlated but separate goals. A truly P2P execution model is proposed, under which the logical execution of a collaborative CQP consists of a set of individual peer CQP executions run by the participating parties. These peer executions are based on the same CQP template; but each peer represents a designated role, it actively processes the steps belong to that role, and skip the steps not belonging to that role. They synchronize peer CQP executions by passing the query return messages. This collaborative process execution model represents a shift from coordinator based distributed query processing to truly P2P collaboration.

This work represents an initial step to the synthesis of relational operation, UDF and business process. We further envisage that the proposed P2P CQP execution model will have impact not only on inter-enterprise collaboration but also on the cooperation of multiple share-nothing data servers.