WiFi, LTE, or Both? Measuring Multi-Homed Wireless Internet Performance Shuo Deng, Ravi Netravali, Anirudh Sivaraman, Hari Balakrishnan MIT CSAIL
Motivation How good are LTE and WiFi networks in practice? How do they compare with each other? Should a mobile device use one or both networks?
Main Findings No clear cut for WiFi and LTE selection Crowd-Sourcing Measurement Multi-Path TCP not always outperform TCP Interface selection for primary subflow matters Controlled Measurement on Multi-Path TCP Short-flow apps benefit most from TCP Mobile App Measurement
Main Findings No clear cut for WiFi and LTE selection Crowd-Sourcing Measurement Multi-Path TCP not always outperform TCP Interface selection for primary subflow matters Controlled Measurement on Multi-Path TCP Short-flow apps benefit most from TCP Mobile App Measurement
Crowd-Sourcing Measurement Free App for Android ( >1000 downloads) Measures WiFi and LTE performance on the same device at (almost) the same time Signal Strength Ping DNS Query Time TCP Throughput …
Crowd-Sourcing Measurement World-wide data collection: 16 countries, over 3000 measurements
Crowd-Sourcing Measurement 20% of the time, LTE has lower RTT than WiFi
Crowd-Sourcing Measurement 40% of the time, LTE has higher TCP throughput than WiFi Uplink Downlink 8
Crowd-Sourcing Summary WiFi and LTE can win over each other significantly Mobile devices should be able to choose between WiFi and LTE adaptively Or, use both?
Main Findings No clear cut for WiFi and LTE selection Crowd-Sourcing Measurement Multi-Path TCP not always outperform TCP Interface selection for primary subflow matters Controlled Measurement on Multi-Path TCP Short-flow apps benefit most from TCP Mobile App Measurement
Multi-Path TCP Setup Measurement setup Measured at 20 different locations on the east and west coasts of the United States
MPTCP performs worse than TCP Multi-Path TCP vs TCP 33% of the time, Multi-Path TCP gives higher throughput than TCP MPTCP performs worse than TCP
Multi-Path TCP Measurement Measure Multi-Path TCP using different configs Use WiFi or LTE for primary subflow Use coupled or decoupled congestion control Use backup and non-backup mode Quantify by relative difference |Tput1–Tput2|/Tput1
Multi-Path TCP Primary Subflow Using different interface for primary subflow has a significant impact on short flows
Multi-Path TCP Congestion Ctrl Using different congestion control algorithms affects long flows more
Multi-Path TCP Backup Mode Energy efficiency: MPTCP may consume excessive energy, even in Backup Mode LTE, Non-backup Mode LTE, Backup Mode WiFi, Non-backup Mode WiFi, Backup Mode
Multi-Path TCP Summary Multi-Path TCP does not always outperform TCP Choosing the proper interface for primary subflow improves MPTCP throughput for short flows Using different congestion control affects long flow more Multi-Path TCP may consume excessive amount of energy
Takeaways Multi-homed devices need to choose between WiFi and LTE adaptively Devices need to decide when to use Multi-Path TCP or TCP Multi-Path TCP does not always outperform TCP Multi-Path TCP may not be energy efficient Devices need to decide which network to transmit on first when using Multi-Path TCP