Current NCDesk Icon NCDesk

Technical Notes - Bandwidth Footprint Analysis

Results of analysis conducted averaging all forms of the spring 2005 field test on July 20 & 21, 2005

Communications during the test
Packets
Bytes
https received
954
145597
https sent
1052
179560
https total
2006
325157
http received
6
545
http sent
6
457
http total
12
1002
Total received
960
146142
Total sent
1058
180017
Total sent and received
2018
326159
Total Footprint
326.16
Kb
Total data rate (send and receive) over 90 minutes 
.48
Kbps

Almost as much data is transferred to the server as received from the server. Averaging out the data transmitted over 90 minutes, results in a data rate of .27 Kbps, sent upstream. Since most ISPs bias traffic flow against traffic going upstream, this should be the rate we are most concerned about. This is a very small data rate. Overall, the test requires transmition of a minimal amount of data.

The graph below shows that traffic tends to burst up to 10,000 Bytes/10 secs, or 8 Kbps, when bandwidth is freely available in the network. Bursts half that size are more common. The largest packets are those containing encrypted application data.

To get an idea how many users may take the test at once, given no other applications traffic on the network, assume the baseline is the 56K modem. While the paragraph above implies that seven users should be able to take the test at once on a 56K modem, in practice this is an unreasonable expectation. This is due to the fact that computers generate overhead traffic, and ISP's tend to overbook bandwidth. From practical experience, only one or two users may take the test over a 56K modem.

Extrapolating support for concurrent test takers assuming no other network traffic on the following connections:
Connection
Users
Notes
ISDN
4-8
 
DSL
13-26
( if the pipe is symmetric)
Cable Modem
6-12
(based on an asymmetric pipe biased in favor of downloads)
T1
18-36
 

Please note that keeping to the minimum limits above will yield very good performance in general. As you start to approach the upper limit, performance will degrade, such that the test will be slow at the upper limits. Also, if other applications run at the same time as the test on the network, performance will be negatively impacted.

This file was modified on Wednesday, August 29, 2007; at 1:36:41 PM