To pipeline or not to pipeline, that is the question

TL;DR

This paper clarifies the concepts of pipelining and blocking in query processing, introduces a spectrum based on unit-of-transfer, and provides an analytical model showing the narrow performance gap between these methods in in-memory databases.

Contribution

It defines a clear terminology for data transfer in query pipelines, introduces a spectrum of techniques, and develops an analytical model for performance analysis in in-memory systems.

Findings

The gap between pipelining and non-pipelining performance is narrow.

A spectrum of data transfer techniques exists based on unit-of-transfer.

Designers should reconsider the traditional pipelining vs. blocking dichotomy.

Abstract

In designing query processing primitives, a crucial design choice is the method for data transfer between two operators in a query plan. As we were considering this critical design mechanism for an in-memory database system that we are building, we quickly realized that (surprisingly) there isn't a clear definition of this concept. Papers are full or ad hoc use of terms like pipelining and blocking, but as these terms are not crisply defined, it is hard to fully understand the results attributed to these concepts. To address this limitation, we introduce a clear terminology for how to think about data transfer between operators in a query pipeline. We show that there isn't a clear definition of pipelining and blocking, and that there is a full spectrum of techniques based on a simple concept called unit-of-transfer. Next, we develop an analytical model for inter-operator communication, and highlight the key parameters that impact performance (for in-memory database settings). Armed with this model, we then apply it to the system we are designing and highlight the insights we gathered from this exercise. We find that the gap between pipelining and non-pipelining query execution, w.r.t. key factors such as performance and memory footprint is quite narrow, and thus system designers should likely rethink the notion of pipelining vs. blocking for in-memory database systems.