HybridSFC: Accelerating Service Function Chains with Parallelism

Yang Zhang, Zhi Li Zhang, Bo Han

Research output: Chapter in Book/Report/Conference proceedingConference contribution

1 Scopus citations

Abstract

Network Function Virtualization (NFV) coupled with Software Defined Networking (SDN) creates new opportunities as well as substantial challenges such as increased Service Function Chain (SFC) latency and reduced throughput. In this paper, we present HybridSFC, a framework that explores the opportunities of parallel packet processing at both traffic level and Network Function (NF) level. It incorporates innovative control and data-plane mechanisms that partition and convert a sequential chain into several (finer-grained) SFClets that can be executed in parallel on multiple cores and servers. HybridSFC is practical in that it can handle NFs spanning multiple servers and requires no modifications to existing NFs. Experiments show that HybridSFC reduces latency up to 51% with 7% CPU overhead, and a 1.42×-1.87× improvement in overall system throughput.

Original languageEnglish (US)
Title of host publicationIEEE Conference on Network Function Virtualization and Software Defined Networks, NFV-SDN 2019 - Proceedings
EditorsLarry Horner, Kurt Tutschku, Fabrizio Granelli, Yuji Sekiya, Marco Tacca, Deval Bhamare, Helge Parzyjegla
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728145457
DOIs
StatePublished - Nov 2019
Event2019 IEEE Conference on Network Function Virtualization and Software Defined Networks, NFV-SDN 2019 - Dallas, United States
Duration: Nov 12 2019Nov 14 2019

Publication series

NameIEEE Conference on Network Function Virtualization and Software Defined Networks, NFV-SDN 2019 - Proceedings

Conference

Conference2019 IEEE Conference on Network Function Virtualization and Software Defined Networks, NFV-SDN 2019
CountryUnited States
CityDallas
Period11/12/1911/14/19

Fingerprint Dive into the research topics of 'HybridSFC: Accelerating Service Function Chains with Parallelism'. Together they form a unique fingerprint.

Cite this