Mobile Access Bandwidth in Practice: Measurement, Analysis, and Implications

Table of Contents

Introduction

Data Release

Implementation of Cross-Layer and Cross-Technology Measurement Tool

Implementation of New Bandwidth Testing Service (Swiftest)

Correction of Bugs in Our Paper

Introduction

Our study focuses on characterizing mobile access bandwidth in the wild. We work with a major commercial mobile bandwidth testing (BTS) app to analyze mobile access bandwidths of 3.54M end users based on fine-grained measurement and diagnostic information collected by our cross-layer and cross-technology measurement tool. Besides, our analysis provides insights on building an ultra-fast, ultra-light bandwidth testing service (BTS) called Swiftest at scale. Swiftest dramatically reduces the test time of the commercial BTS app from 10 $sec$ to 1 $sec$ on average, with a 15$\times$ reduction on the backend cost. This repository contains our implementation of the cross-layer and cross-technology measurement tool, our released data, and our implementation of Swiftest.

Data Release

The involved data include

1) a four-month fine-grained dataset (collected by us),

2) a two-year coarse-grained dataset (provided by BTS-APP operation team).

Currently, we have released a portion of the representative sample data (with proper anonymization) belonging to the former dataset for references, including the fine-grained collected data of 50K tests in 4G, 5G, and WiFi 4, 5, 6 access technologies (10K tests each). These data are organized in 4G.csv, 5G.csv, wifi4.csv, wifi5.csv, and wifi6.csv, respectively (for detailed data, please click here). For each CSV file, we list the specific information coupled with the regarding description as follows. As to the remainder, we are still in discussion with the authority to what extend can it be released.

4G.csv

Column Name	Description
user_uid	Unique ID generated to identify a user (cannot be related to the user’s true indentity)
brand	Mobile device brand
model	Mobile device model
network_type	Network type
os_version	Android version
user_isp_id	ISP id
user_region_id	Region (province) of a user
user_city_id	City of a user
bandwidth_Mbps	Bandwidth testing result in Mbps
cell_asuLevel	LTE signal strength in ASU value
cell_dbm	LTE signal strength in dBm
cell_level	Abstract level value for the overall signal quality
cell_mcc	3-digit mobile country code
cell_mnc	2 or 3-digit mobile network code
cell_mobileNetworkOperator	5 or 6-digit code (MCC+MNC) for mobile network operator
cell_rssi	Received signal strength indication (RSSI) in dBm
cell_timingAdvance	Timing advance value for LTE
cell_bands	Bands of the LTE connection
cell_bandwidth	Cell bandwidth in kHz
cell_ci	28-bit cell identity code
cell_earfcn	18-bit absolute radio frequency channel node
cell_tac	16-bit Tracking Area Code
cell_rsrp	Reference signal received power in dBm
cell_rsrq	Reference signal received quality
cell_rssnr	Reference signal signal-to-noise ratio

5G.csv

Information	Description
user_uid	Unique ID generated to identify a user (cannot be related to the user’s true indentity)
brand	Mobile device brand
model	Mobile device model
network_type	Network type
os_version	Android version
user_isp_id	ISP id
user_region_id	Region (province) of a user
user_city_id	City of a user
bandwidth_Mbps	Bandwidth testing result in Mbps
cell_asuLevel	LTE signal strength in ASU value
cell_dbm	LTE signal strength in dBm
cell_level	Abstract level value for the overall signal quality
cell_mcc	3-digit mobile country code
cell_mnc	2 or 3-digit mobile network code
cell_bands	Bands of the LTE connection
cell_tac	16-bit Tracking Area Code
cell_nrarfcn	New radio absolute radio frequency channel number
cell_csiRsrp	CSI reference signal received power
cell_csiRsrq	CSI reference signal received quality
cell_ssRsrp	SS reference signal received power
cell_ssRsrq	SS reference signal received quality
cell_ssSinr	SS signal-to-noise and interference ratio

wifi4/wifi5/wifi6.csv

Information	Description
user_uid	Unique ID of a user (cannot be related to the user’s true indentity)
brand	Mobile device brand
model	Mobile device model
network_type	Network type
os_version	Android version
user_isp_id	ISP id
user_region_id	Region (province) of a user
user_city_id	City of a user
bandwidth_Mbps	Bandwidth testing result in Mbps
wifi_rssi	The received signal strength indicator of the current 802.11 network in dBm
wifi_linkSpeed	Current link speed
wifi_hiddenSSID	Whether this network does not broadcast its SSID
wifi_frequency	Current WiFi frequency in MHz
wifi_rxLinkSpeedMbps	Current receive link speed in Mbps
wifi_txLinkSpeedMbps	Current transmit link speed in Mbps
wifi_maxSupportedRxLinkSpeedMbps	Maximum supported receive link speed in Mbps
wifi_maxSupportedTxLinkSpeedMbps	Maximum supported transmit link speed in Mbps
wifi_wifiStandard	Connection Wi-Fi standard

Implementation of Cross-Layer and Cross-Technology Measurement Tool

We have released the project of our cross-layer and cross-technology (CLCT) measurement tool here. Note that this project can be directly compiled and run using the Android Studio platform with the support of Java SE Development Kit 8.

Implementation of New Bandwidth Testing Service (Swiftest)

Currently we are scrutinizing the codebase to avoid possible anonymity violation. To this end, we will release Swiftest’s source code in a module-by-module manner as soon as we have finished examining a module and acquire its release permission from the authority. The codebase of Swiftest is organized as follows.

Swiftest
|---- client-side
|---- server-side
          |---- test-server
          |---- master-server

• Swiftest/client-side is an Android Studio project similar to the CLCT measurement tool.
• Swiftest/server-side/test-server currently includes a simplified version of test servers’ transmission logic. We are still negotiating with BTS-APP’s operational team to acquire the release permission of the entire transmission logic of test servers.
• Swiftest/server-side/master-server contains the source code (in Go) and the executable binary of master-server.

The released part can be found here.

Correction of Bugs in Our Paper

After the publication of our paper, we have found out some bugs in the published version under the help of several warm-hearted researchers. We list the details below.

Location	Original Version	Corrected Version	Bug Type	Contributor
Section 3.3, last paragraph	In comparison, for 4G networks, we find that the average bandwidth at different times of the days is in general positively correlated with the number of tests conducted by users.	In comparison, for 4G networks, we find that the average bandwidth at different times of the days is in general negatively correlated with the number of tests conducted by users.	Typo	Lei Zhang from UCAS

Manys thanks to all the contributors’ help and contributions!