Loading...
Thumbnail Image
Item

Algorithmic mapping of software-defined networking pipelines

Abstract
Networks with a consistent software stack reduce the complexity of monitoring, testing and automation, and reduce the mental burden on operators. However, when network software is bundled with the hardware, operators face being locked into a single vendor’s hardware, when they might otherwise be able to use cheaper or more suitable hardware from another vendor. Ostensibly, Software Defined Networking (SDN) gives operators the freedom to operate hardware from different vendors using the same software stack. However, producing SDN software that controls hardware from different vendors is hampered by differences in the packet processing pipelines of the ASICs each vendor uses. This thesis presents the design and evaluation of Shoehorn, a system for improving the portability of SDN control-plane software. Shoehorn finds mappings from virtual pipelines (defining the packet processing requirements of control-plane software), to physical pipelines (defining the packet processing pipeline of a physical device). Shoehorn improves on current approaches by ensuring that the mappings are suitable for real-time translation of controlchannel instructions, by ensuring a one-to-one mapping of virtual pipeline table entries to physical pipeline table entries. This also ensures that the mappings do not significantly increase the memory usage or power consumption of the pipelines. This thesis evaluates Shoehorn by mapping 25 virtual pipelines, based on real SDN control-plane software for managing diverse networks, to a variety of physical pipelines, based on real hardware SDN implementations. The evaluation finds that all but 6 virtual pipelines are supported by multiple physical pipelines, and that in every case where Shoehorn could not find a mapping, it was due to a virtual table that no table in the physical pipeline could support.
Type
Thesis
Type of thesis
Series
Citation
Date
2022
Publisher
The University of Waikato
Rights
All items in Research Commons are provided for private study and research purposes and are protected by copyright with all rights reserved unless otherwise indicated.