|
The
test network consisted of two Cisco 2621 routers, a Cisco 7206VXR
router, and two Cisco 3548 switches that were set up in order to
emulate two service provider networks: a content provider network,
and a single user. Instead of using an independent web server we
used a built in web server on the PathControl box for testing. (See
diagram above.)
The
7206 used default BGP attributes in selecting the route used to
get to the single user on the remote network. (Step 1) Next, in
order to influence the PathControl’s ability to make egress
routing decisions, we inserted two Linux servers configured as IP
forwarders. A network impairment tool called nistnet (available
free from http://snad.ncsl.nist.gov/itg/nistnet)
was installed on each of the Linux servers to provide a controlled
degradation of service by allowing configurable latency and frame
loss to be added to the network path.
Using
nistnet, we introduced 300ms of latency and 25% frame loss into
the default BGP preferred path. The web client then requested a
page that was served of the PathControl’s management module.
While our control router, a Cisco 7206 router, still preferred the
degraded connection due to the way BGP selects the preferred route,
the PathControl determined that an alternate route existed through
the path to the other service provider.
PathControl
made this determination by measuring the HRTT between its USTAT
modules and to the end user across both ISP connections. Since the
PathControl device was configured to assert routes, it advertised
the preferred route by adding a higher weight to the route through
the other ISP to the 7206 router. When the latency and packet loss
was removed from the original path, the PathControl adjusted the
preferred route back to the original default egress path on the
7206 router.
|
