>
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...
media,4
Since the time when our initial experiments were run, the vBNS
connection to the University of Washington has been activated.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... media.5
At the time of these experiments, CalTech's vBNS connection had been
approved.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... application.6
The only difference is that the experimental data server outputs to
a socket, whereas the production server runs as a CGI-bin script
(and therefore outputs to standard output).
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... times.7
Comparison experiments were run back-to-back and multiple times on
transfer sizes ranging from 1.5 to 3 megabytes (corresponding to a
typical SARA user request). Each comparison run executed for
between 15 seconds and 2 minutes with the majority of runs being on
the lower end of the range. On average, comparison runs for a
single trial enjoyed roughly similar load conditions.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... re-established.8
When this phenomenon was first noticed, we began running
traceroutes to each machine to determine if network traffic
at any given time was using the vBNS. In one instance, when the NWS
reported a sudden increase in bandwidth, traceroute
indicated that the vBNS link was being used. Closely monitoring the
link, we found that approximately two hours later, the routing
changed from vBNS to the general Internet, accompanied by a sudden
decrease in bandwidth.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.