NAP Architecture

Anyone willing to speak forth and give notes of their experience in
setting up a NAP.

Questions we have in mind are:

Go with a Digital GIGAswitch/ATM or FDDI?
Or should we go for a GSR 12000?

The very high traffic users will be put directly on ports on the
GIGAswitch or GSR. The others will be sharing FDDI.


  I expect there will be several folks who will be at
  this NANOG who might be able to answer. Whether they
  are willing is another matter. At previous NANOGs
  there have been some discussions on exchange architectures
  and scalability tradeoffs. You might find some good
  stuff in the archives.

The choice of hardware is really dependant on quite a bit. Obviously one
needs a highly reliable infrastructure. Also one will want to choose
something that has the ability to grow with the exchange points needs.

Additionally, current high traffic exchange points are at least showing
some of the faults in the hardware they are currently using.

However, I think at least today, one could generalize, and come to the
following conclusions.

If you are looking to create an exchange point where peering is forced,
(i.e. a central routing model), You could choose to use some high
performance router. The problems one may run into here is port cost and
administrative costs. Additionally, the current batch of super-high
performance routers are fairly new, and one may end up spending alot of
time fighting bugs.

If you are going to go with the more common bridged type of exchange point,
you have a fairly wide choice of hardware, including FDDI, Fast Ethernet,
ATM, etc. Problems with all these technologies are well documented, and
there are several exchange operators who are using at least one of these
technologies, and would best have the detailed answers to your questions.
However, it might be of relevence to note that the Internet 2 project has
chosen ATM as the technology of choice for the MEGA and GIGA pops of the
new network. With the detail of information one must provide to recieve
federal funding, you might be able to dig up some good analysis from an
Internet 2 submittal.

Chris A. Icide