Join us for an exciting talk on GPU-accelerated databases with Bowen Wu from ETH Zurich (Systems Group). He will present Eiger, a cutting-edge query engine built for high-performance analytics. Discover how this work is advancing database efficiency and unlocking new optimizations for modern computing. Don’t miss this opportunity!

Title: Eiger: An efficient GPU-based query engine for data analytics

Speaker：Bowen Wu | ETH Zurich

When: 03/13(Thur) 13:00-14:00 (UTC+00:00)London

Where: Online

External: https://app.huawei.com/wmeeting/join/98061574/rVE2nkYLUbuzVUIxNytixOuK4Kj6bpj3q

Meeting ID: 98061574

Passcode: 626816

Registration: https://www.smartsurvey.co.uk/s/3N8U7J/

Abstract:

There is a growing interest in leveraging GPUs for tasks beyond ML, especially in database systems. For this reason, we are building a GPU-based query engine for OLAP workloads called Eiger. It features a collection of highly efficient GPU-based relational operators, a query optimizer, the cross-platform ability to run on both NVIDIA and AMD GPU platforms, and multi-GPU support. In this talk, I will mainly focus on our effort to optimize GPU-based operators efficiently in Eiger, which has been accepted to SIGMOD 2025. Despite the existing extensive work on GPU-based database operators, several questions are still open. For instance, the performance of almost all operators suffers from random accesses, which can account for up to 75% of the runtime. In addition, the group-by operator, which is widely used in combination with joins, has not been fully explored for GPU acceleration. Furthermore, existing work often uses limited and unrepresentative workloads for evaluation and does not explore the query optimization aspect, i.e., how to choose the most efficient implementation based on the workload. In this paper, we revisit the state-of-the-art GPU-based join and group-by implementations. We identify their inefficiencies and propose several optimizations. We introduce GFTR, a novel technique to reduce random accesses, leading to speedups of up to 2.3x. We further optimize existing hash-based and sort-based group-by implementations, achieving significant speedups (19.4x and 1.7x, respectively). We also present a new partition-based group-by algorithm ideal for high group cardinalities. We analyze the optimizations with cost models, allowing us to predict the speedup. Finally, we conduct a performance evaluation to analyze each implementation. We conclude by providing practical heuristics to guide query optimizers in selecting the most efficient implementation for a given workload. In addition to operator implementations, I’ll also briefly mention Eiger’s cross-platform and multi-GPU capabilities.

Bio: 

Bowen Wu is a PhD candidate in Computer Science at ETH Zürich, where he is advised by Professors Gustavo Alonso. His research focuses on optimizing databases for heterogeneous computing platforms, particularly GPU-accelerated query processing. Prior to his PhD, he earned his Master’s degree with distinction from ETH Zürich and a First Class Honors Bachelor’s degree from The Chinese University of Hong Kong. Bowen has interned at Microsoft Gray Systems Lab, working on distributed multi-GPU platforms, and at AWS, contributing to query optimization for Amazon Redshift.

【BAYES COFFEE HOUSE TECH TALK SERIES】Eiger: An efficient GPU-based query engine for data analytics / Huawei-Edinburgh Joint Lab by blogadmin is licensed under a Creative Commons Attribution CC BY 3.0