
Traffic Splitting

Sites have come up with a number of different techniques for dividing
their commodity (Internet) traffic from their research (vBNS)
traffic. Here we outline several techniques that people are using, or
are proposing to use, by way of examples.
- Separate circuits.
Utilizing separate infrastructure for connections to the vBNS and
connections to the Internet is the easiest way to split your traffic.
- SONET layer MUXing.
This is just as easy as utilizing separate circuits. However,
this method allows sites to purchase local loop, and
possibly even long-haul access circuits at "bulk" discount prices.
The same SONET infrastructure carries all of the site's traffic, but
the traffic is carried on logically separate circuits, allowing for
simplicity in configuration and management.
- ATM layer MUXing.
Here, a single ATM connection carries both commodity and research
traffic to the site. The two types of traffic are segregated onto
separate ATM Virtual Circuits (VCs). These VCs can be configured with
rate-limits which permit precise control of how much traffic is
permitted on both the commodity and research connections.
- IP layer MUXing.
In this case, traffic is shared at the IP layer. Note that the
connection might still be a ATM connection; however, separate VCs are
not in use. The single IP connection carries traffic from the campus
to an external point where traffic is split (such as a Gigapop). In
this case, both research and commodity traffic are permitted to burst
to the total amount of available bandwidth on the connection. Several
new technologies are becoming available which allow for controlling
the separation of bandwidth between commodity and research traffic.
These include Weighted Random Early Detection (WRED) (available for
Cisco routers), Class Based Queuing (being developed for Ascend
Gigarouter), and the IETF Resource Reservation Protocol (and
associated Integrated Services standards) which are being developed
for a range of platforms.