DPFS: DPU-Powered File System Virtualization

Peter Jan Gootzen; Jonas Pfefferle; Radu Stoica; Animesh Trivedi

doi:10.1145/3579370.3594769

DPFS: DPU-Powered File System Virtualization

Peter Jan Gootzen, Jonas Pfefferle, Radu Stoica, Animesh Trivedi

Research output: Chapter in Book / Report / Conference proceeding › Conference contribution › Academic › peer-review

Abstract

As we move towards hyper-converged cloud solutions, the efficiency and overheads of distributed file systems at the cloud tenant side (i.e., client) become of paramount importance. Often, the clientside driver of a cloud file system is complex and CPU intensive, deeply coupled with the backend implementation, and requires optimizing multiple intrusive knobs. In this work, we propose to decouple the file system client from its backend implementation by virtualizing it with an off-the-shelf DPU using the Linux virtio-fs software stack. The decoupling allows us to offload the file system client execution to a DPU, which is managed and optimized by the cloud provider, while freeing the host CPU cycles. DPFS, our proposed framework, is 4.4× more host CPU efficient per I/O, delivers comparable performance to a tenant with zero-configuration and without modification of their host software stack, while allowing workload and hardware specific backend optimizations. The DPFS framework and its artifacts are publically available at https://github.com/IBM/DPFS.

Original language	English
Title of host publication	SYSTOR '23: Proceedings of the 16th ACM International Conference on Systems and Storage
Publisher	Association for Computing Machinery, Inc
Pages	1-7
Number of pages	7
ISBN (Electronic)	9781450399623
DOIs	https://doi.org/10.1145/3579370.3594769
Publication status	Published - Jun 2023
Event	16th ACM International Conference on Systems and Storage, SYSTOR 2023 - Haifa, Israel Duration: 5 Jun 2023 → 7 Jun 2023

Conference

Conference	16th ACM International Conference on Systems and Storage, SYSTOR 2023
Country/Territory	Israel
City	Haifa
Period	5/06/23 → 7/06/23

Bibliographical note

Funding Information:
We would like to thank our shepherd Youyou Lu and our anonymous reviewers for their valuable feedback. We would also like to thank Nvidia for providing us the BlueField-2 prototype firmware that provided virtio-fs support. Animesh Trivedi is supported by the Dutch Research Council (NWO) grant number OCENW.KLEIN.561.

Publisher Copyright:
© 2023 Owner/Author(s).

Funding

We would like to thank our shepherd Youyou Lu and our anonymous reviewers for their valuable feedback. We would also like to thank Nvidia for providing us the BlueField-2 prototype firmware that provided virtio-fs support. Animesh Trivedi is supported by the Dutch Research Council (NWO) grant number OCENW.KLEIN.561.

Funders	Funder number
Nederlandse Organisatie voor Wetenschappelijk Onderzoek	OCENW.KLEIN.561
Nederlandse Organisatie voor Wetenschappelijk Onderzoek

Keywords

cloud
datacenter
DPU
file system
framework
FUSE
NFS
offloading
RDMA
SmartNIC
storage
virtio-fs
virtualization

Access to Document

10.1145/3579370.3594769

Persistent URL (handle)

Persistent link

Cite this

@inproceedings{0aa2cbc8c3e242b4a21e5422fc98c745,

title = "DPFS: DPU-Powered File System Virtualization",

abstract = "As we move towards hyper-converged cloud solutions, the efficiency and overheads of distributed file systems at the cloud tenant side (i.e., client) become of paramount importance. Often, the clientside driver of a cloud file system is complex and CPU intensive, deeply coupled with the backend implementation, and requires optimizing multiple intrusive knobs. In this work, we propose to decouple the file system client from its backend implementation by virtualizing it with an off-the-shelf DPU using the Linux virtio-fs software stack. The decoupling allows us to offload the file system client execution to a DPU, which is managed and optimized by the cloud provider, while freeing the host CPU cycles. DPFS, our proposed framework, is 4.4× more host CPU efficient per I/O, delivers comparable performance to a tenant with zero-configuration and without modification of their host software stack, while allowing workload and hardware specific backend optimizations. The DPFS framework and its artifacts are publically available at https://github.com/IBM/DPFS.",

keywords = "cloud, datacenter, DPU, file system, framework, FUSE, NFS, offloading, RDMA, SmartNIC, storage, virtio-fs, virtualization",

author = "Gootzen, {Peter Jan} and Jonas Pfefferle and Radu Stoica and Animesh Trivedi",

note = "Funding Information: We would like to thank our shepherd Youyou Lu and our anonymous reviewers for their valuable feedback. We would also like to thank Nvidia for providing us the BlueField-2 prototype firmware that provided virtio-fs support. Animesh Trivedi is supported by the Dutch Research Council (NWO) grant number OCENW.KLEIN.561. Publisher Copyright: {\textcopyright} 2023 Owner/Author(s).; 16th ACM International Conference on Systems and Storage, SYSTOR 2023 ; Conference date: 05-06-2023 Through 07-06-2023",

year = "2023",

month = jun,

doi = "10.1145/3579370.3594769",

language = "English",

pages = "1--7",

booktitle = "SYSTOR '23: Proceedings of the 16th ACM International Conference on Systems and Storage",

publisher = "Association for Computing Machinery, Inc",

}

Gootzen, PJ, Pfefferle, J, Stoica, R & Trivedi, A 2023, DPFS: DPU-Powered File System Virtualization. in SYSTOR '23: Proceedings of the 16th ACM International Conference on Systems and Storage. Association for Computing Machinery, Inc, pp. 1-7, 16th ACM International Conference on Systems and Storage, SYSTOR 2023, Haifa, Israel, 5/06/23. https://doi.org/10.1145/3579370.3594769

DPFS: DPU-Powered File System Virtualization. / Gootzen, Peter Jan; Pfefferle, Jonas; Stoica, Radu et al.
SYSTOR '23: Proceedings of the 16th ACM International Conference on Systems and Storage. Association for Computing Machinery, Inc, 2023. p. 1-7.

Research output: Chapter in Book / Report / Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - DPFS

T2 - 16th ACM International Conference on Systems and Storage, SYSTOR 2023

AU - Gootzen, Peter Jan

AU - Pfefferle, Jonas

AU - Stoica, Radu

AU - Trivedi, Animesh

N1 - Funding Information: We would like to thank our shepherd Youyou Lu and our anonymous reviewers for their valuable feedback. We would also like to thank Nvidia for providing us the BlueField-2 prototype firmware that provided virtio-fs support. Animesh Trivedi is supported by the Dutch Research Council (NWO) grant number OCENW.KLEIN.561. Publisher Copyright: © 2023 Owner/Author(s).

PY - 2023/6

Y1 - 2023/6

N2 - As we move towards hyper-converged cloud solutions, the efficiency and overheads of distributed file systems at the cloud tenant side (i.e., client) become of paramount importance. Often, the clientside driver of a cloud file system is complex and CPU intensive, deeply coupled with the backend implementation, and requires optimizing multiple intrusive knobs. In this work, we propose to decouple the file system client from its backend implementation by virtualizing it with an off-the-shelf DPU using the Linux virtio-fs software stack. The decoupling allows us to offload the file system client execution to a DPU, which is managed and optimized by the cloud provider, while freeing the host CPU cycles. DPFS, our proposed framework, is 4.4× more host CPU efficient per I/O, delivers comparable performance to a tenant with zero-configuration and without modification of their host software stack, while allowing workload and hardware specific backend optimizations. The DPFS framework and its artifacts are publically available at https://github.com/IBM/DPFS.

AB - As we move towards hyper-converged cloud solutions, the efficiency and overheads of distributed file systems at the cloud tenant side (i.e., client) become of paramount importance. Often, the clientside driver of a cloud file system is complex and CPU intensive, deeply coupled with the backend implementation, and requires optimizing multiple intrusive knobs. In this work, we propose to decouple the file system client from its backend implementation by virtualizing it with an off-the-shelf DPU using the Linux virtio-fs software stack. The decoupling allows us to offload the file system client execution to a DPU, which is managed and optimized by the cloud provider, while freeing the host CPU cycles. DPFS, our proposed framework, is 4.4× more host CPU efficient per I/O, delivers comparable performance to a tenant with zero-configuration and without modification of their host software stack, while allowing workload and hardware specific backend optimizations. The DPFS framework and its artifacts are publically available at https://github.com/IBM/DPFS.

KW - cloud

KW - datacenter

KW - DPU

KW - file system

KW - framework

KW - FUSE

KW - NFS

KW - offloading

KW - RDMA

KW - SmartNIC

KW - storage

KW - virtio-fs

KW - virtualization

UR - http://www.scopus.com/inward/record.url?scp=85165993519&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85165993519&partnerID=8YFLogxK

U2 - 10.1145/3579370.3594769

DO - 10.1145/3579370.3594769

M3 - Conference contribution

AN - SCOPUS:85165993519

SP - 1

EP - 7

BT - SYSTOR '23: Proceedings of the 16th ACM International Conference on Systems and Storage

PB - Association for Computing Machinery, Inc

Y2 - 5 June 2023 through 7 June 2023

ER -

DPFS: DPU-Powered File System Virtualization

Abstract

Conference

Bibliographical note

Funding

Keywords

Access to Document

Persistent URL (handle)

Other files and links

Fingerprint

Cite this