Designing a Fast, Efficient, Cache-friendly Hash Table, Step by Step

Play Designing a Fast, Efficient, Cache-friendly Hash Table, Step by Step
Sign in to queue

Description

Hash tables consume a large volume of both compute resources and memory across Google's production system. The design for hash tables in C++ traces its origins to the SGI STL implementation from 20 years ago. Over these years, computer architecture and performance has changed dramatically and we need to evolve this fundamental data structure to follow those changes. This talk describes the process of design and optimization that starts with std::unordered_map and ends with a new design we call "SwissTable", a 2-level N-way associative hash table. Our implementation of this new design gets 2-3x better performance with significant memory reductions (compared to unordered_map) and is being broadly deployed across Google.

Day:

3

Session Type:

Presentation

Code:

102

Embed

Download

Right click to download this episode

Download captions

The Discussion

Add Your 2 Cents