Mobile and Cellular Networks from the Edge

Sat, 01 Nov 2014 00:00:00 +0000

Project Overview

For most people on Earth, the Internet is reached over a cellular network. Yet cellular infrastructure is among the hardest of all networks to observe: the interesting behaviour happens inside operator networks that publish little, use private addressing, and terminate traffic at gateways whose location bears no simple relation to the user’s.

The consistent move in this thread is to measure from the client, where a handset can see what an outside prober cannot, and to keep asking a question that outsiders usually assume away: for a mobile user, where is the network actually, and what does that cost?

Cellular Networks in the Internet

Cell Spotting mapped where cellular traffic enters the Internet, showing that the gateways serving mobile users are far more concentrated and far less local than the geography of their users implies. Behind the Curtain followed the consequence into content delivery: if a CDN sees a query arriving from a distant gateway, the replica it picks is wrong for the user holding the phone — the same problem our DNS work had found on fixed lines, sharper in cellular.

Anycast on the Move asked what happens to anycast routing when the client itself is moving, and Where Things Roam turned to cellular IoT and M2M connectivity, a population of devices that roams permanently by design.

Aggregators and Roaming

Roaming is no longer a bilateral arrangement between two operators. A layer of global aggregators now sits between the SIM and the networks that serve it, which changes the performance a device gets and makes the path it takes hard for anyone to predict. Our MobiSys paper laid out a taxonomy of these aggregators and measured their roaming performance and optimization, and the later TNSM article extends the comparison.

The Mobile Web

Mobile web performance has been shaped as much by intermediaries as by protocols. AMP up your Mobile Web Experience characterized what Google’s Accelerated Mobile Pages actually delivers, separating the gains attributable to the format from those attributable to serving from Google’s cache. Reining in Mobile Web Performance looked at what document and permission policies can recover, and Mobile AD(D) examined app session times, which determine what an ad — or any load — is worth paying for.

Connectivity in Motion

In-flight Internet is a useful extreme case: a satellite path, a shared cell with a hundred passengers, and a network stack designed for none of it. Mile High WiFi was a first look at in-flight connectivity in the wild, and When IPs Fly made the case for rethinking airline communication rather than treating the aircraft as an ordinary subscriber.

Papers

Cellular networks in the Internet

Behind the Curtain – Cellular DNS and Content Replica Selection — IMC 2014
Behind the Curtain: The Importance of Replica Selection in Next Generation Cellular Networks — SIGCOMM 2014 (poster)
Cell Spotting: Studying the Role of Cellular Networks in the Internet — IMC 2017
Anycast on the Move: A Look at Mobile Anycast Performance — TMA 2018
Where Things Roam: Uncovering Cellular IoT/M2M Connectivity — IMC 2020

Aggregators and roaming

Global Mobile Network Aggregators: Taxonomy, Roaming Performance and Optimization — MobiSys 2022
A Comparative Analysis of Global Mobile Network Aggregators — IEEE TNSM, 2025

The mobile web

Mobile AD(D): Estimating Mobile App Session Times for Better Ads — HotMobile 2015
AMP up your Mobile Web Experience — MobiCom 2019
Reining in Mobile Web Performance with Document and Permission Policies — HotMobile 2022

Connectivity in motion

When IPs Fly: A Case for Redefining Airline Communication — HotMobile 2016
Mile High WiFi: A First Look at In-Flight Internet Connectivity — WWW 2018

Coverage

In-flight WiFi — Northwestern McCormick News, ChicagoInno, and the Chicago Tribune (2016)
The in-flight browsing extension — Network World, New Atlas, ScienceDaily, CACM News, Indian Express, and Travel+Leisure (2017)
Phone usage — Northwestern McCormick News and ACM TechNews (2016)

People

Lead & PI

Fabián E. Bustamante (Northwestern University)

Students

Collaborators

Andra Lutu (Telefónica Research)
Sergi Alcalá-Marín, Marcelo Bagnulo (Universidad Carlos III de Madrid)
Özgü Alay (University of Oslo)
Moritz Steiner (Akamai)
Vishnu Navda, Ranjita Bhagwan, Saikat Guha (Microsoft Research India)

Measurement Platforms for Mobile Devices

Wed, 01 Feb 2012 00:00:00 +0000

Project Overview

Studying mobile networks and applications runs into a practical wall: the device. A phone will not let you sit in the middle of its traffic, its battery behaviour cannot be simulated convincingly, and the person holding it has their own plans. Each of those obstacles produced an instrument.

These are tools rather than findings — the apparatus that the measurement studies in our mobile thread depend on.

MBZ: a middlebox on the handset

Traditional middleboxes see traffic at the far end of the first link, by which point the interesting attribution — which app, which library, which permission — is gone. MBZ takes that vantage point and moves it onto the device: an extensible platform that intercepts and can modify traffic before it leaves the handset, or on arrival before it reaches the operating system.

Conceptually it gives a middlebox’s view with the device’s context attached, which is what makes it possible to ask questions about specific applications rather than about aggregate flows. Back in Control describes the design.

BatteryLab: power measurement on real hardware

Energy claims about mobile software are usually made with models, because measuring real draw needs physical instrumentation attached to a real device. BatteryLab is a collaborative platform that shares that instrumentation: devices wired to power monitors, hosted across sites, available remotely so researchers can measure genuine power consumption on hardware they do not own.

Crowd (Soft) Control

The third obstacle is the human. Opportunistic sensing takes measurements wherever participants happen to be, which leaves coverage to chance. Crowd (Soft) Control asked what becomes possible when participants can be gently steered toward where a measurement is needed, and the later HotMobile paper takes up the question of incentives — what actually motivates people to participate, and why one scheme does not fit every population.

Software

Middle-Box Zero (MBZ) — an extensible platform to intercept and modify traffic on the device, before it leaves or before it reaches the OS. GitHub
BatteryLab — a distributed platform for power monitoring of mobile devices, led by Matteo Varvello. Its project site no longer resolves; see the PAM 2022 paper below.

Papers

Crowd (Soft) Control — HotMobile 2012
No One-Size Fits All: Towards a Principled Approach for Incentives in Mobile Crowdsourcing — HotMobile 2014
Back in Control: An Extensible Middle-Box on Your Phone — arXiv, 2020
BatteryLab: A Collaborative Platform for Power Monitoring — PAM 2022

Coverage

Crowd Soft Control was covered by Communications of the ACM, Wired, and Discovery News in 2012.

People

Lead & PI

Fabián E. Bustamante (Northwestern University)

Students

John P. Rula
James Newman

Collaborators

Matteo Varvello (Nokia Bell Labs)
Kleomenis Katevas, Mihai Plesa, Hamed Haddadi (Brave Software / Imperial College London)
Narseo Vallina-Rodriguez (IMDEA Networks)
Abbas Razaghpanah (Stony Brook University)
Mark Allman (ICSI)

Mobile Networks | Aqualab - Northwestern University

Mobile and Cellular Networks from the Edge

Project Overview

Cellular Networks in the Internet

Aggregators and Roaming

The Mobile Web

Connectivity in Motion

Papers

Coverage

People

Related Links

Measurement Platforms for Mobile Devices

Project Overview

MBZ: a middlebox on the handset

BatteryLab: power measurement on real hardware

Crowd (Soft) Control

Software

Papers

Coverage

People

Related Links