WiFi-Reports: Improving Wireless Network Selection Jeffrey Pang (CMU) with Ben Greenstein (IRS) Michael Kaminsky (IRP) Damon McCoy (U. Colorado) Srinivasan Seshan (CMU)

Problem: Commercial AP Selection tmobile attwifi (ap 1) attwifi (ap 2) seattlewifi linksys Free Public Wifi $3.99 $9.99 Free! Which networks will run my applications? Which ones have good performance? Quality = ??? We often have many choices of wireless access points (APs), but little information about each Jiwire.com Hotspot database Jiwire.com Hotspot database

Wifi-Reports Database Bandwidth: 30 kbps Blocked ports: Bandwidth: 5 Mbps Blocked ports: None community uploads measurement reports members download summary statistics

Wifi-Reports tmobile attwifi (ap 1) attwifi (ap 2) seattlewifi linksys Free Public Wifi I need to use VoIP so this is the best network for me Bandwidth: 300 kbps Blocked ports: None Doesn’t work! Wifi-Reports provides information about AP performance and application support Doesn’t work! Bandwidth: 100 kbps Blocked ports: None Bandwidth: 300 kbps Blocked ports: None Wifi-Reports Hotspot database Wifi-Reports Hotspot database Bandwidth: 30 kbps Blocked ports: Bandwidth: 5 Mbps Blocked ports: None Doesn’t work!

Research Challenges Doesn’t work! Bandwidth: 300 kbps Blocked ports: None Location privacy – Reports shouldn’t be linked, otherwise they can be used to track users – But also need to limit fraud; e.g., 1 report per AP per user – Solution: new ecash-like reporting protocol & robust summary functions

Research Challenges Doesn’t work! Bandwidth: 300 kbps Blocked ports: None Bandwidth: 10 Mbps Works great! (Really!) Location privacy – Reports shouldn’t be linked, otherwise they can be used to track users – But also need to limit fraud; e.g., 1 report per AP per user – Solution: new ecash-like reporting protocol & robust summary functions

Research Challenges Location privacy – Reports shouldn’t be linked, otherwise they can be used to track users – But also need to limit fraud; e.g., 1 report per AP per user – Solution: new ecash-like reporting protocol & robust summary functions Location context – Performance dependent on location with respect to AP – Wireless channel effects loss rate – Solution: estimate different loss regimes w/ distributed measurements Doesn’t work! Bandwidth: 300 kbps Blocked ports: None Bandwidth: 10 Mbps Works great! (Really!)

WiFi-Reports Overview WiFi-Reports Account Service WiFi-Reports Account Service Independent Report Databases Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on UBookstore Cafe: Bandwidth: 4000 kbps Blocked ports: , Skype,… Report on UBookstore Cafe: Bandwidth: 4000 kbps Blocked ports: , Skype,… Users collect measurement reports when they use networks Reports are sent to databases for others to download

Questions and Challenges How useful would this system be in practice? – Do users have many choices of usable wireless networks? – Is there diversity in network performance/functionality? – Is performance stable enough to be predictive? – Are networks better for some applications but worse for others? – Do enough users use real networks to gather measurements? How do we limit “report fraud” and provide anonymity? – Not even the account service should be able to link a user’s reports to each other (otherwise location privacy is violated) – But each user should only be able to report once per network

Part I: Measurement Study We built a measurement tool to simulate reports that users would collect – Bandwidth, latency, jitter, blocked ports, number of other users, etc. We measured all networks visible from: – 13 hotspots near The Ave – ~7 days at different times of day Measurement procedure: – Sit near center of hotspot – Perform active spot measurements (2-3 minutes per network) shinka tea tullys 1 starbucks 1 tullys 2 trabant oasis lounjin yunnie bubble tea sureshot bookstore cafeontheave starbucks 2 cafesolstice Our study is the first to examine pay networks and encrypted networks, in addition to open networks Our study is the first to examine pay networks and encrypted networks, in addition to open networks

Results: Are there many APs? Better

Results: Are there many APs? Better

Results: Is there diversity? Better

Results: Is there diversity? Better

Results: Are measurements predictive? Better

Results: Are there application trade-offs? Better Better latency Better bandwidth

Part II: Private and Accountable Reporting How do we limit “vote fraud” and provide anonymity? Requirements: – No one, even the account service, should be able to link a user’s reports to each other (otherwise previous work says location privacy is violated) – Each user should only be able to report once per network

Anonymizing Mix Network Design Sketch WiFi-Reports Account Service WiFi-Reports Account Service Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Independent Report Databases tmobile seattlewifi CAFEONTHEAVE … 1.Client creates one token per AP 2.Client blinds each token 3.Account Service signs blinded tokens 4.Client unblinds tokens … 1.Client uses and measures an AP 2.Client uses token to sign report 3.Report is published via mix network Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None Report on tmobile: Bandwidth: 1200 kbps Blocked ports: None

Design Sketch Details = random public key pair {K tmobile, K tmobile -1 } – Account Service signs the public key K tmobile – Private key K tmobile -1 is used to sign reports  can update reports Assumptions: – The account service only gives one identity to each user – Most users are honest Open issues: – Client must get all tokens signed before using them  expensive If only get signed token after AP is used, use of the AP is revealed Can obtain a subset of all tokens instead (e.g., all tokens for a city)  trade off between token signing overhead for more location privacy

Token Generation Time Time to generate tokens for all APs in one city (single threaded Xenon 3.4 Ghz server)

Resistance to Fraud CDF of Prediction accuracy (1 = most accurate)

Ongoing work Implemented Wifi-Reports service – Linux client exists – Currently implementing Android client. Anyone want to help? MobiSys ‘09

Questions? (Other summer projects)

Results: Is there diversity? Better Blocked port = no measurement

Results: Is there diversity? Better Blocked port = no measurement

Results: Are measurements predictive? Better Blocked port = no measurement