Characterizing Deep-Learning I/O Workloads in TensorFlow

TL;DR

This paper analyzes TensorFlow's I/O performance, identifies bottlenecks, and proposes a burst buffer solution to significantly improve checkpointing efficiency and overall training performance.

Contribution

It provides a detailed characterization of TensorFlow's I/O behavior and introduces a burst buffer method to enhance checkpointing performance.

Findings

Increasing threads boosts bandwidth up to 7.8x.

Prefetching overlaps computation and I/O, eliminating I/O costs.

Burst buffer improves checkpointing speed by 2.6x.

Abstract

The performance of Deep-Learning (DL) computing frameworks rely on the performance of data ingestion and checkpointing. In fact, during the training, a considerable high number of relatively small files are first loaded and pre-processed on CPUs and then moved to accelerator for computation. In addition, checkpointing and restart operations are carried out to allow DL computing frameworks to restart quickly from a checkpoint. Because of this, I/O affects the performance of DL applications. In this work, we characterize the I/O performance and scaling of TensorFlow, an open-source programming framework developed by Google and specifically designed for solving DL problems. To measure TensorFlow I/O performance, we first design a micro-benchmark to measure TensorFlow reads, and then use a TensorFlow mini-application based on AlexNet to measure the performance cost of I/O and checkpointing in TensorFlow. To improve the checkpointing performance, we design and implement a burst buffer. We find that increasing the number of threads increases TensorFlow bandwidth by a maximum of 2.3x and 7.8x on our benchmark environments. The use of the tensorFlow prefetcher results in a complete overlap of computation on accelerator and input pipeline on CPU eliminating the effective cost of I/O on the overall performance. The use of a burst buffer to checkpoint to a fast small capacity storage and copy asynchronously the checkpoints to a slower large capacity storage resulted in a performance improvement of 2.6x with respect to checkpointing directly to slower storage on our benchmark environment.