1998-08-17 - Re: Internet is rickety

Header Data

From: Bill Stewart <bill.stewart@pobox.com>
To: Alex Alten <cypherpunks@toad.com
Message Hash: 7087aa2164aae7b1aabeb7295237417917849472f3f01640966edd0c79dcb9b4
Message ID: <3.0.5.32.19980816232208.00bc7510@idiom.com>
Reply To: <199808102320.SAA12845@mixer.visi.com>
UTC Datetime: 1998-08-17 06:22:52 UTC
Raw Date: Sun, 16 Aug 1998 23:22:52 -0700 (PDT)

Raw message

From: Bill Stewart <bill.stewart@pobox.com>
Date: Sun, 16 Aug 1998 23:22:52 -0700 (PDT)
To: Alex Alten <cypherpunks@toad.com
Subject: Re: Internet is rickety
In-Reply-To: <199808102320.SAA12845@mixer.visi.com>
Message-ID: <3.0.5.32.19980816232208.00bc7510@idiom.com>
MIME-Version: 1.0
Content-Type: text/plain


This isn't coderpunks material; I'm copying to cypherpunks instead.
(I've also removed letters@sjmercury.com from the Cc:.)

At 08:01 AM 8/16/98 -0700, Alex Alten wrote:
>Bruce,
>You are making it sound worse than it is.  The underlying network 
>routing infrastructure is designed to be extremely robust in the face 
>adverse conditions (originally it was nuclear attack).  I'm only aware 
>of two incidents that actually ground large parts of the Internet to a 
>halt.  The 1st involved a bug in the backbone rounters when the Internet 
>was young and had relatively few nodes.  The 2nd was the Morris worm 

Physical damage is one thing; logical-level damage is another.
Some newer vulnerabilities are the DNS system and routing protocols.

There was a major DNS failure a year or so ago when NSI or one of the
root servers published a garbled database; I don't remember details.
That's annoying, but easily noticed, and easy to work around,
and easier to prevent happening in the future.  But it's also easy
to trick the DNS root servers from outside (as Kashpureff did),
or mess with them inside (I forgot why Postel did that recently.)

A more serious problem occurred when some small ISP misconfigured
their Border Gateway Protocol parameters, announcing to the whole world
that their little T1 line was the best way to get almost anywhere
from almost anywhere else.  This was only mildly untrue until a
few large providers updated their tables to reflect it, at which point
their T1 started to melt, taking much of the Internet down with it.
This kind of thing _can_ happen again; the interesting problem
is what if somebody deliberately starts doing it a lot....

And much of the world's traffic goes through MAE-WEST and MAE-EAST,
as a way of getting between ISPs, though bigger carriers
have direct connections between their networks, and many small ISPs
at least have direct connections to their continent-wide carriers.
The MAEs and other NAPs get slow fairly often, and occasionally
just get fried; a major outage could be a serious problem,
so the fact that MAE-EAST lives in a parking garage in the DC area
may or may not bother you.

That's three ways that the whole Internet can have serious trouble;
there are presumably a number of others, if you believe in Murphy
or believe the L0pht's announcements.

Then there are more subtle problems - what happens if Altavista dies?
(Yes, there are other search engines, but probably 3-5 major ones.)
And there are the increasing number of portal and free email services from
Hotmail, Excite, Altavista, MyYahoo, Netscape, etc.,
which will take out large numbers of users' mail access if any of them dies.
And then there's AOL, which is probably a pretty robust design,
and which may or may not be missed by people on the _real_ Internet
if it dies for a while, but could also cut off lots of users from the net.

Also, much of the robustness derives from having _enough_ excess bandwidth
in the network that if you lose a good chunk of it, you'll still have
enough for the remaining users (the nuclear war scenario does reduce
the user population as well as the network...)  But you can get into
localized problems, for example the cable cut that trashed uunet's access 
to Florida the other week, and the Internet doesn't have a widespread priority
mechanism to identify "more important" traffic, though there are
many mechanisms for prioritizing individual users' networks.

It still works pretty well, but that doesn't mean it's not rickety.

				Thanks! 
					Bill
Bill Stewart, bill.stewart@pobox.com
PGP Fingerprint D454 E202 CBC8 40BF  3C85 B884 0ABE 4639





Thread