Network Status Traces on Multi-Homed Devices

Experiment Overview

The objective of the CoSphere Trial is to collect network availability and network status traces as observed on multi-homed devices of real users. We ran this experiment with 12 users in a six week period in February/March 2007, logging cellular, 802.11 wireless lan, Bluetooth and (indirectly) fixed USB events on Windows CE devices. These web pages provide an overview of the collected data and a first step analysis.

We want to use the traces of this experiment to investigate the possibilities for 'experience-based network resource usage on mobile hosts' (see poster). The basic idea is that knowledge on the mobility patterns of the user helps to optimize the usage of network resources by applications on mobile hosts, especially for those applications that have a degree of delay-tolerance. We have focused on the prediction in time of a mobility event of interest, such as 'getting in range of the home 802.11 network'. We propose a method using preceding events as predictors for the event of interest, as discussed in the papers Density Estimation for Out-of-Range Events on Personal Mobile Devices and Predicting mobility events on personal devices.

This work is part of the CoSphere PhD project.

The Feb/Mar 2007 experiment uses version 1.1 of the CoSphere NAL software. Preparation and execution in cooperation with SocioXensor.

Basic Numbers

Duration (for 11 participants):

Network entities (for 12 participants):

On the x-axis are the participant identifiers. The black lines are the per participant average and the red lines are the globally unique number of the parameter under consideration divided by the number of partipicants. The closer the two lines, the less overlap in network entity visibility between users.

Measurements

The mobile devices of the participants have multiple network interfaces: a GPRS/GSM cellular interface, an 802.11 wireless lan interface, a Bluetooth interface and a fixed USB interface. The trial software logs changes in the status of these network interfaces and initiates periodical activation and scanning. Furthermore, the software keeps track of the IP configuration.

The following network resource information is stored:

• Cellular: cellid / lac of currently used base station
• Cellular: name and number (incl. country code) of in-range operators
• WLAN: addresses of in-range 802.11 access points
• WLAN: name and set of access points of in-range 802.11 networks
• Bluetooth: name and address of in-range nodes
• IP: IP configuration of available interfaces

Intervals:

• Cellular: allways on with cellid status updates through asynchronous notifications
• Cellular: operator list obtained every 5 minutes
• WLAN: activated 1 minute every 10 minutes, which triggers scanning and association automatically
• Bluetooth: always on and discoverable with an inquiry scan every 5 minutes
• IP: asynchronous notification upon changes

The WLAN interface can not be permanently on because of its high power consumption. We have tried to tune the software such that the partipicants only needed to recharge their devices once a day. The XML log files together are several hundred MBytes large.