Back to search

Cornell University

Towards Scalable and Optimal Oblivious Reconfigurable Networks

Abstract

dc:description.abstract

Datacenter network demands show explosive growth, doubling nearly every year. Unfortunately, datacenter networks are built primarily using packet switches, which do not scale as quickly as these demands and are expected to scale even slower in the future. Nanosecond-scale optical circuit switches represent a potential alternative: unlike packet switches, they are not limited by semiconductor scaling trends, and unlike previous optical circuit switches, they are fast enough to support all datacenter traffic types, including short flows. To be used to their full potential, however, these switches will require novel network designs. This dissertation examines how to build datacenter-scale networks using exclusively nanosecond-scale optical circuit switches. We identify the Oblivious Reconfigurable Network (ORN) design paradigm, which is designed to use the capabilities of these switches. We develop Shale, the first ORN to achieve a tunable tradeoff between latency scalability and throughput. We also show how to compose multiple tunings to support multiple traffic classes, a common feature of datacenter network traffic. To enable Shale, we develop a novel congestion control algorithm tailored to Shale’s unique environment, which we extend to address node and link failures. Finally, we implement a Field-Programmable Gate Array (FPGA)-based hardware prototype for a Shale end-host. Our designs show that Shale can achieve orders of magnitude better latency and hardware resource requirements than previous ORN designs. Additionally, we investigate the fundamental performance limits of ORNs, and prove that ORNs must grapple with an inherent tradeoff between latency and throughput. The tradeoffs achieved by Shale match this fundamental tradeoff up to a constant factor, meaning that every tuning of Shale is Pareto optimal among ORNs. Together, Shale and our exploration of the fundamental limits of ORNs represent important steps towards scalable and optimal ORNs.

Degree

thesis:*
Name thesis:degree_name
Ph. D., Computer Science
Level thesis:degree_level
Doctor of Philosophy
Discipline thesis:degree_discipline
Computer Science
Grantor
Cornell University
Year dc:date.issued
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Amir, Daniel
Chair dc:contributor.chair
  • Weatherspoon, Hakim
Committee members dc:contributor.committeemember
  • Kleinberg, Robert
  • Campana, Andrew

Subjects

dc:subject × 4

Rights

dc:rights
Statement dc:rights
  • Attribution 4.0 International
Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
16611696
Dc Identifier Other
ProQuest Submission ID: 14581
ProQuest Publication ID: 31488702

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Amir, Daniel. Towards Scalable and Optimal Oblivious Reconfigurable Networks. Doctor of Philosophy thesis, Cornell University, 2024. https://hdl.handle.net/1813/116381