 |
The direction from the local computer (where SProbe runs) to a remote Internet machine is called the downstream direction. To measure the bottlleneck bandwidth of the downstream direction, a generic SProbe command is: |
sprobe -p 100 -s 1460 -t 6 -c 10 -d machine_to_measure
-p indicates the port number. Try to pick a non-responsive port (port 100 seems
to be a consistently good choice)
-s indicates payload size of the probing packets. These packets should be large.
1460 is the Ethernet frame -- most of people should not change this
-t indicates the number of probing packets (out of which only 2 are large). Use
either 6, 4 or 2. Typically 6 is a good idea.
-c indicates the number of times SProbe tries to produce an estimate. (In
certain cases - cross-traffic - SProbe chooses not to report an estimate).
-d indicates the direction - downstream
Example: To measure bottleneck
bandwidth to www.cs.berkeley.edu, I use:
./sprobe -p 100 -s 1460 -t 6 -c 10 -d www.cs.berkeley.edu
Note that hosts behind firewalls
will appear to SProbe as being offline. Most high visibility web servers are
behind firewalls (e.g. www.cnn.com).
Also, make sure you don't send large
streams of SYN packets (small t, please). Be warned that SYN streams can raise
alarms on some firewalls.
|
The direction from a remote Internet machine to the local computer is called the upstream direction. In this direction, SProbe can only measure bottleneck bandwidths from Web servers and Gnutella peers.
-p is the port number - for a web server the port is usually 80 Example: To measure bottleneck bandwidth from www.cnn.com, I use: ./sprobe -p 80 -h www.cnn.com
sprobe -p 6346 -g ip_addr_of_Gnutella_host
-p is the port number Example: To measure bottleneck bandwidth from 128.95.3.93, I use: ./sprobe -p 6346 -g 128.95.3.93
|
