Picks and Pans

Wow boy. I'm not sure. I have a cable modem. I love it. But
it has failed. And when it did they said I would have to wait
a week to have it fixed. My heart sank. I get so much email
that this would cause a serious problem. My mail server can
not be off-line for a week.

So, I installed an ISDN backup system. Luckily this is easy for me.
For many people this would be hard/impossible/expensive. The good new
is that my cable modem came back to life the next day. It turns out
that for a while it was flakey when it rained (I now understand this
is not an uncommon phenomena for cable modems).

Posted by Brad at 12:00 AM | Permalink | TrackBacks (0)

OK, so I have my cable modem, personal domain, mail server, etc...
I even back up my files to tape. I discovered recently I'm my
own personal mini-ISP to myself. How scary.

So then, I discover that my packets are taking 1/4 second to get
from my house to work when I can drive there in 15 minutes. This
might see OK but if you do the math and remember how fast the
speed of light is, it should take a lot less than 1/4 second or
250 milliseconds (we say "250ms" in the trade).

So, I do some probing (it turns out that 250ms is *very* noticeable
if you are typing in a telnet session) and found out that my packets
went from Boston to San Francisco and then back to Boston then to my
office. Nothing like a little trip to the coast and back. This
little "mis-routing" was cause by a mis-configuration at a place
known as MAE-east. A "MAE" is a metropolitan area Ethernet. This
is an idea cooked up by a company know as MFS and now known as WorldCom.
Loosely it's a way to connect up a bunch of Ethernet speed LAN's
using fiber and routers. Think of it as a room in a building near
Washington DC with lots of routers and blinking lights. Anyway,
in this room many (but not all) ISP's connect their networks together.

When they connect them together they need a way to communicate who
and what is where on the combined network. These days we use
"routing protocols" to let routers speak to each other about what
is where. SO, if the routing protocol says the wrong thing, or
if one router misunderstands or misinterprets the routing protocol
information, your packets go to San Francisco instead of Billerica.

OK, that's complex stuff. Imaging if you understand it, but don't
work for a ISP. Imagine trying to explain to them that there is
a routing problem. I did this for 2 (yes, two) months before I got
my cable company to fix the problem. Ugg. Even the phone company
was not that slow. The eventually fixed the problem (one ISP
was emitting a BGP-4 MED which cause them to prefer a route to
MAE-west over a local route to MAE-east, for the routing geeks) and
now my packets take only 40ms to get to work (that's 5 times faster
and I am much happier).

I think the cable companies are much more at home dealing with
people who can not get to the ESPN web site, but this is complex
stuff at the bottom and not fully baked yet...

Posted by Brad at 12:00 AM | Permalink | TrackBacks (0)

Starting up new leased line Internet connections is pain. It's
pain for the users and a pain for the ISP. It's slow, error prone
and manual labor intensive.

For leased lines it would be nifty if the ISP would run PPP and
force the remote side to "log in". Once logged in it seems like
the ISP router could use RADIUS to assign the IP address to the
line and add any static routes. The router would then have to
'remember' the config and the associated leased line connection.
The pay back is that turning up new lines would be automagic and only
require a human if it did not work...

It would require that the remote router be able to identify itself
with some sort of "secret", but that's not hard. The identity should
also be tied to billing information so the 'back end' systems at
the ISP can start billing for the line when the initial log-in
occurs..

So what? Why is this important? Well, if the CLEC's (a nifty
acronym of the new wave of phone-company-like companies which
can rent copper lines from the existing RBOCs) appear like everyone
seems to think they will, and they start selling T1 lines to
mortals like you and me for $60/month, they won't be able to
turn the lines on fast enough to satisfy the demand. (sorry for
that run on - I got carried away).

Some folks believe that $60 T1 lines will cause explosive growth.
If one of the barriers to growth is how fast one can provision
new lines, automated provisioning will become very important.

I think there's a new protocol here, or at least some extensions
to RADIUS.

Posted by Brad at 12:00 AM | Permalink | TrackBacks (0)