A comparison of two high-speed connections

by Moo and Me
moo and me at dynamicnetservices dot com

Introduction

We're interested in determining which high-speed internet connection is faster, cable or DSL.

Materials

Two high-speed internet services were piped into the same downtown Toronto (Rosedale) location, 3Web 10 Megabit high-speed cable (Gold package) over Roger's cable network, and Teksavvy 5 Megabit residential high-speed DSL (premium edition) over Bell Canada's DSL network. Two "gateway" servers were used, one for each internet connection. Both servers ran the Slackware 12.0 operating system, with a typical network configuration. An Acanac high speed online personal computer VPS (Extreme package) was used to provide the download bandwidth used for this experiment. The Acanac virtual server is capable of a sustained upload rate of 20 Megabits per second.

Procedure

The download speeds were monitored for almost five days. Download speeds were measured by timing how long each gateway took to download a test file from the dedicated high-bandwidth server. The test file contained 14,035,988 bytes of compressed Linux log files. The cron utility was used to execute a bash script that measured the download speeds every half an hour. Download speeds were measured using three different protocols, FTP, HTTP, and SCP. The commands used in the bash script were:

time -f "FTP %e" wget ftp://TheAcanacServer.com /test --user=?? --password="??" --progress=dot -r
time -f "HTTP %e" wget http://TheAcanacServer.com/test --progress=dot -r
time -f "SCP %e" scp -q ??@TheAcanacServer.com:test .

Other than the test file downloads, there was no significant network traffic on the gateways during the five days of the test, and in particular, there was no peer-to-peer communications traffic for the duration of the test.

Results

No significant difference between the download speeds of the three protocols was observed.

The average speeds reveal a large drop in throughput for the 3Web cable connection in the evening hours, see Table 1 and Figure 1.




Table 1 - Average download speeds observed in different time ranges, in kilobytes per second.

Connecton   
24 hours
   (i.e. overall average)   
   5:00pm - 3:00am   
   10:00pm - 1:00am   
3Web cable modem   
604
387
166
Teksavvy DSL   
473
468
465




Figure 1 - Average download speeds versus time, in kilobytes per second.


(Click for larger image)




An Excel spreadsheet containing the raw data is available here.

Conclusions

The Teksavvy DSL connection consistently provided the promised bandwidth throughout the entire day. The throughput of the 3Web cable connection, on the other hand, fluctuated greatly throughout the day. In particular, during the block of time that most people are likely to be home, awake, and using a high speed internet connection (5:00pm to 3:00am), slightly less than half (387 kilobytes per second) of the peak bandwidth (840 kB/s) was available.

During prime time (10:00pm to 1:00am), the Teksavvy DSL connection averaged 465 kB/s, while the 3Web cable connection dropped to 166 kB/s.

During non-primetime hours, the 3Web cable connection provided higher average (758 kB/s) and peak (840 kB/s) download speeds than the Teksavvy DSL connection (average non-primetime: 477 kB/s, peak non-primetime: 500 kB/s).

The better service for a particular subscriber depends upon the time of day that the subscriber uses the service. Those who predominantly use their home internet connection after work, or after dinner, and are in bed by 1:00am, will get almost three times the bandwidth from Teksavvy's DSL, than 3Web's cable service.

Future Work

During the pilot study prior to this experiment, we noticed that the total download speed through the 3Web cable connection could be increased by performing more than one download simultaneously. This may suggest that the dramatic drop in bandwidth through 3Web is a result of bandwidth throttling instead of network congestion. Further investigation is necessary.