First Responder: Persistent Memory Simultaneously as High Performance Buffer Cache and Storage

Authors: 

Hyunsub Song, Shean Kim, J. Hyun Kim, Ethan JH Park, and Sam H. Noh, UNIST

Abstract: 

Persistent Memory (PM) is a new media with favorable characteristics that can vastly improve storage I/O performance. While new PM based file systems have been developed to exploit PM, most work have not been successful in fully integrating PM media with traditional storage media such as SSDs and HDDs. We present First Responder (FR), a means to exploit the beneficial features of PM, while making use of modern and mature file systems such as Ext4 developed for traditional storage devices. Conceptually, FR is much like a buffer cache, but much more is involved such as maintaining consistency under failure and providing featherweight management overhead. FR brings about multiple benefits. First, we retain the maturity of existing file systems allowing deployment of FR at settings where traditional file systems are deployed. Second, traditional storage devices supported by these file systems can be used allowing easy integration of PM with traditional storage. Finally, FR allows in-order file system semantics at close to PM device latency. With experimental evaluations with the Intel DC PMM, we show that FR, when used in cache form, can outperform Ext4 by more than 9×, while providing durable in-order file system semantics, whereas Ext4 cannot. We also show that when used as part of a typical file system, performance is comparable with NOVA and Ext4-DAX.

Open Access Media

USENIX is committed to Open Access to the research presented at our events. Papers and proceedings are freely available to everyone once the event begins. Any video, audio, and/or slides that are posted after the event are also free and open to everyone. Support USENIX and our commitment to Open Access.

BibTeX
@inproceedings {273928,
author = {Hyunsub Song and Shean Kim and J. Hyun Kim and Ethan JH Park and Sam H. Noh},
title = {First Responder: Persistent Memory Simultaneously as High Performance Buffer Cache and Storage},
booktitle = {2021 USENIX Annual Technical Conference (USENIX ATC 21)},
year = {2021},
isbn = {978-1-939133-23-6},
pages = {839--853},
url = {https://www.usenix.org/conference/atc21/presentation/song},
publisher = {USENIX Association},
month = jul
}

Presentation Video