A case for disaggregation of ML data processing

Metadata

Presented in arxiv:2210.14826.

Authors: Andrew Audibert, Yang Chen, Dan Graur, Ana Klimovic, Jiri Simsa, Chandramohan A Thekkath (Google & ETH Zurich)

Understanding the paper

TL;DR

This paper present tf.data service, which is a disaggregated input data processing service built on top of tf.data.

Features

• Disaggregate data preprocessing from ML computation.

• Horizontal scaling.

• Data sharing to reuse computation (e.g., hyperparameter tuning jobs).

• Coordinated data reads.

Evaluation

• Speedups: up to 100x.

• Reduce job cost: up to 89x.