What is it about?
Data centers are embracing the disaggregated memory (DM) architecture to improve resource efficiency and save costs. Many types of range indexes, such as B+ trees, radix trees, and learned indexes, are proposed to store and retrieve data on DM, but they cannot simultaneously achieve low cache consumption and low read amplifications. This paper proposes a hybrid index as a better choice and optimizes it to fit DM.
Featured Image
Photo by Kevin Ache on Unsplash
Why is it important?
Today's cloud data centers have long suffered from low memory utilization (< 60%). Disaggregated memory (DM) is an increasingly prevalent architecture proposed to address such an issue. However, traditional indexing solutions do not work well on either I/O performance or cache efficiency when ported to DM, which are equally important.
Perspectives
This data structure is specifically designed for disaggregated memory. It also contains some useful and novel designs, such as a fully one-sided RDMA-based implementation for hopscotch hash tables and a general sibling-based validation design for B+ trees. I hope you find this paper interesting.
Xuchuan Luo
Fudan University
Read the Original
This page is a summary of: CHIME: A Cache-Efficient and High-Performance Hybrid Index on Disaggregated Memory, November 2024, ACM (Association for Computing Machinery),
DOI: 10.1145/3694715.3695959.
You can read the full text:
Resources
Contributors
The following have contributed to this page
