External Memory Pipelining Made Easy With TPIE

TL;DR

This paper introduces an extension to the TPIE library that simplifies the development of I/O-efficient external memory algorithms by enabling pipelining and parallelization, reducing disk I/O overhead and improving performance.

Contribution

The paper presents a major extension to TPIE that adds pipelining, parallelization, and automatic memory management for more efficient external memory algorithm implementation.

Findings

Pipelining reduces I/O overhead in external memory algorithms.

The extended TPIE library supports parallel internal memory computation.

The framework has been successfully used in research and commercial applications.

Abstract

When handling large datasets that exceed the capacity of the main memory, movement of data between main memory and external memory (disk), rather than actual (CPU) computation time, is often the bottleneck in the computation. Since data is moved between disk and main memory in large contiguous blocks, this has led to the development of a large number of I/O-efficient algorithms that minimize the number of such block movements. TPIE is one of two major libraries that have been developed to support I/O-efficient algorithm implementations. TPIE provides an interface where list stream processing and sorting can be implemented in a simple and modular way without having to worry about memory management or block movement. However, if care is not taken, such streaming-based implementations can lead to practically inefficient algorithms since lists of data items are typically written to (and read from) disk between components. In this paper we present a major extension of the TPIE library that includes a pipelining framework that allows for practically efficient streaming-based implementations while minimizing I/O-overhead between streaming components. The framework pipelines streaming components to avoid I/Os between components, that is, it processes several components simultaneously while passing output from one component directly to the input of the next component in main memory. TPIE automatically determines which components to pipeline and performs the required main memory management, and the extension also includes support for parallelization of internal memory computation and progress tracking across an entire application. The extended library has already been used to evaluate I/O-efficient algorithms in the research literature and is heavily used in I/O-efficient commercial terrain processing applications by the Danish startup SCALGO.