Atlanta-NAP's choice of switch

>} Which begs a question: why use a Giga-switch at all?
>} With the head of line blocking feature/problem and scalability only to
>} full duplex 100 mbps is a Gigaswitch something that should be used in a
>} next generation NAP?

Nathan mentioned that MFS has started their new MAEs with a Catalyst or
shared FDDI ring. Perhaps that has something to do with inital demand.
An example of this is MAE-Houston or MAE-LA, neither of which presently
require the bandwidth a Gigaswitch delivers. MFS has been sticking to the
plan of adding hardware and/or capacity based on demand and traffic stats.

I really need to start reading responses to posts before posting
myself. :slight_smile:

I think the Atlanta NAP, while probably a good idea, won't run into the
head of line blocking problem in the extremely near future. Looking at
the growth pattern of other exchange points leads me to believe this.

That may or may not be true.

It's my opinion that an Atlanta based NAP makes much more sense than
the existing MAE-LA nap. There isn't a major public NAP in the
southeastern part of the United States, and it's quite possible that
if done properly, all of the Major NSPs will want to go there for a
common place to peer (even taking into account their propensity to
move towards private peering). If they go there, I know we will.

If you add in the typical 6 month bandwidth doubling trend it leads
directly to my second opinion that it's not going to be enough
bandwidth next year. The trend is real. If you are using up a FDDI
link today, chances are you will be using 4 of them a year from now.

Given the above, the amount of traffic could approach 100 Mbps on an
outbound link quite rapidly, even for the small, mid size NSPs.

>} I'm not suggesting it's intended to be the next generation NAP, but
>} you'd think that they would want to use the latest switches and
>} technology available, rather than continue down the FDDI road.

What else would you suggest? Gigabit Ethernet hasn't been standardized
yet, Cisco doesn't make a HIPPI interface, and some people prefer to not
use ATM. FDDI has proven to very reliable, etc. Having ISPs continue to
grow egress bandwidth has shown to be a bigger problem than the switch
fabric at the larger exchange points.

Sticking my neck out, I would suggest ATM. It scales better today.
Rather than the current trend of running multiple FDDI links to a
switch to handle that egress (and ingress) bandwidth, you can use a
single 0C3/OC12 link and leave the load balancing across multiple
links problems behind.

As for reliability, AADS has shown that you can build an ATM based NAP
that is reliable. Yes there was a failure the other day, but there
have been failures at the FDDI naps also.