NERD: The **N**eed to **E**xperiment on my **Rotten** **D**SL
The NERD experiment is a very, very small effort to understand why the SBC Yahoo! DSL service provided to my residence is so freaking crappy. The collaboration, centered in Redwood City, CA, has recently collected its first month of data and provides here a preliminary analysis, which may be of interest to a general audience in the Emeral Hills area of Redwood City.
Introduction
SBC Yahoo! DSL is the only DSL company which is capable of providing DSL to the Emerald Hills area of Redwood City, CA. Other companies, such as those used by several of the neighbors of this collaboration, are only capable of providing a few 100 kilobits/second (kb/s) of download speed. SBC Yahoo, however, promises 1.5-3.0 Mb/s but can typically deliver no more than 1.2 Mb/s. This is presumably a combination of phone line quality and distance from the phone exchange.
It was observed early on that within 24 hours of turning on the SpeedStream modem, supplied by SBC, the modem will lock up and no longer provide a viable network connection. Attempts to reset the modem from the firmware do not result in an improvement to the situation. The only successful action is to “power cycle” the appliance; that is, switch it off and then switch it back on again.
Replacing the modem supplied by SBC with an identical modem from the SBC Yahoo DSL starter pack sold at Best Buy provided no change in the problem. Replacing the supplied modem with a much older SpeedStream modem, which can provide only bitrates up to ~380 kb/s, solved the problem but resulted in less performance than was being paid for by the collaboration.
Experimental Overview
In order to automate the power cycling of the modem, an X10 appliance control module is plugged into the wall. The modem is then plugged into the module. The open-source X10 control program, “bottlerocket”, is installed on the webserver. A CRON job checks the network connection every minute, and if the connection is determined to be down it issues an “off”, then “on”, command via bottlerocket to the X10 module. Within a window of two minutes, the modem will automatically power up and provide a connection to the server.
It was early on noticed that there appeared to be more of a need to cycle the modem in the morning than in the afternoon. To test this small-statistics observation, the CRON job was setup to e-mail the system administrator every time the modem had to be cycled. This resulted in a data set consisting of 35 days (June 13-July 18, 2005) of time records.
The data were analyzed using the ROOT data analysis framework, enhanced with the RooFit toolkit. All times from the emails were normalized to hour, where a minute (00-59) is 1/60th of an hour and a second (00-59) is 1/3600th of an hour. The times were then plotted in a histogram of 24 bins, one for each hour of the day. The result is shown below in Figure 1.
FIG. 1: the number of times, for a given hour of the day over a period of 35 days, that the SpeedStream modem needed to be power-cycled.
The data are striking. The need to cycle the modem is significantly enhanced between 8am and noon, falling off to a very small level in the late afternoon. There may also be an enhancement between 8-9pm, though more data are required to determine this.
Results
If the cause of the lock-up were connected to only the quality of the house phone line, we would expect that the need to cycle the modem would be independent of the time of day. However, the data contradicts that expectation. We conclude at this preliminary stage that whatever the cause of the problem, it is strongly time-dependent. It is interesting to note that the time of least need for cycling (midnight-8am) corresponds strongly to the hours during which most people slumber in the local area. The spike in the need to power cycle the modem may therefore be connected to the period of highest use of the local network, suggesting it’s not capable ot handling the local demand.
More data will be collected over the next few months, with further reports to follow.
