Detection and Analysis of Infrastructure Bottlenecks in a Cloud-Centric Internet

See also the Cloud Bottlenecks Funding page.

Project Summary

The COVID-19 pandemic has accelerated the Internet’s shift from a peer-to-peer to a cloud-centric model, making cloud services critical to modern online life. As public clouds evolve to meet increasing demands for low latency and high throughput applications, their infrastructure reaches its limits at network borders. This creates a pressing need to understand how well existing Internet backbone networks support the applications and content served by the cloud. While cloud providers can upgrade their own infrastructures, the economics of operating transit backbone networks result in performance bottlenecks, particularly as traffic to cloud services surges.

The Cloud Bottlenecks project aims to design measurement and analysis tools that will transform our understanding of cloud connectivity performance and reachability, both in the U.S. and globally. Researchers currently lack the ability to measure these bottlenecks at scale or assess their impact on Internet users. The project addresses this gap by developing tools that can identify performance bottlenecks between cloud data centers and the broader Internet infrastructure.

The project is organized into two primary tasks:

  1. Identifying Performance Bottlenecks: Developing new techniques to detect bottleneck links between cloud datacenters and public speed test servers, using active measurements and TCP flow analysis.
  2. Comprehensive Path Analysis: Analyzing bottleneck links through large-scale path measurements from cloud datacenters to the public Internet, including inferring the geographic locations of bottleneck links by determining where paths exit cloud networks.

Intellectual Merit

The intellectual merit of this project stems from its development of novel and scalable methods for conducting accurate performance and topology measurements in cloud connectivity. These methods overcome the cost barriers that have historically prevented such measurements. The data generated by the project will offer a foundation for applying machine learning techniques to analyze network infrastructure, which is currently a significant challenge in this domain.

Broader Impact

The project’s broader impacts extend beyond scientific research. The tools and data developed will be useful for enterprises and application developers deploying into the cloud, providing insights into cloud connectivity bottlenecks. For policymakers, this project can illuminate infrastructure bottlenecks in the U.S. and help identify broadband performance disparities. These insights can inform future investments in public infrastructure, potentially leading to more equitable broadband access. Additionally, the project’s findings and experiences will be incorporated into educational materials, including an undergraduate data science course and research mentorships.

Data Providers

RIPE Network Coordination Centre (RIPE NCC)

Funding Support

National Science Foundation (NSF)

This material is based on research sponsored by the National Science Foundation (NSF) grant CNS-2212241. The views and conclusions contained herein are those of the authors and should not be interpreted as necessarily representing the official policies or endorsements, either expressed or implied, of NSF.

Published
Last Modified