[8799] in cryptography@c2.net mail archive
Re: tapping undersea fibers?
daemon@ATHENA.MIT.EDU (John Denker)
Mon Jun 4 12:53:33 2001
Message-Id: <4.2.2.20010604102318.025f3e30@127.0.0.1>
Date: Mon, 04 Jun 2001 12:09:52 -0400
To: Lenny Foner <foner@media.mit.edu>, cryptography@wasabisystems.com
From: John Denker <jsd@research.att.com>
In-Reply-To: <200106040438.AAA16456@out-of-band.media.mit.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset="iso-8859-1"; format=flowed
Content-Transfer-Encoding: quoted-printable
I wrote:
>>AfricaONE has a backbone that circles the continent offshore, plus=20
>>separate drops for each country, when it would have been vastly cheaper=20
>>to go by land....[Offshore is] less likely to be tapped by hostile powers.
At 12:38 AM 6/4/01 -0400, Lenny Foner wrote:
>My understanding of this, which could very well be wrong, was not
>concern about tapping, but concern over damage.
Well, of course there is concern about all sorts of threats:
*) Tapping
*) Damage
-- inadvertent
-- otherwise
But let's ask what are the costs and benefits of various options:
1) One obvious option is to build a _ring_ of fiber on dry land.
1a) Sure, there will be an inadvertent cut now and then (due to some=20
klutz with a backhoe) but such cuts can be repaired. The cost of repairs=20
is infinitesimal compared to the cost of running the cable=20
offshore. Because it is a ring, customers won't notice cuts if they happen=
=20
one at a time (and are repaired promptly); it would take _two_ cuts to=20
cause a partition.
1b) There will be wars now and then. Each combatant _will_ want to cut=
=20
the adversary's cable. (Ever hear of a foreign minister named=20
Zimmerman?) There's not a lot that the cable operator can do to maintain=20
service in the war zone. But remember it takes two cuts, including one=20
_not_ in the war zone, before anybody outside the war zone is left in the=
dark.
1c) It is not impossible to have damage (inadvertent or otherwise) to=20
an offshore cable, as Herr Zimmerman found out.
1d) A dry-land configuration would provide !much! more functionality --=
=20
more places for paying customers to connect.
2) You can do even better if build something more like a mesh than a=20
simple ring. Such a structure would be tremendously robust against damage,=
=20
and would provide even more places for customers to connect.
So it seems to me that the principal rationale for putting the cable=20
offshore is the expectation that the two-bit warlord next door would have a=
=20
hard time tapping the backbone.
>Is there technology in existence that can usefully encrypt the entire
>contents of such cables' data stream, end-to-end?
It's certainly doable. You might think it would get easier year by year,=20
as electronics gets cheaper and faster -- but fibers are getting faster,=20
too. The crucial factor is that the crypto market has grown to the point=20
where people are actually making stuff for this market, such as 3DES in=20
hardware at (nearly) OC-48 speeds:
http://www.10xinc.com/DES.html
Each wavelength on the AfricaONE fiber is OC-192 (10 Gb/sec). You would=20
have to pick the incoming OC-192 data apart into four or five OC-48=20
streams, encrypt it, re-assemble it, and send it to the laser. That would=
=20
work OK for a point-to-point link, but more generally you would need to add=
=20
"outer headers" =E0 la IPsec. And you might want to worry about key=20
management :-).
-- A full solution would be a nifty piece of engineering. You can't buy=
=20
it at K-mart.
-- But it would be a lot cheaper than putting the cable offshore=20
unnecessarily.
People say that CBC prevents you from parallelizing the encipherment, but=20
in practice it's straightforward to make it work (for modest degrees of=20
parallelism). Or you can use counter mode.
>It seems to avoid a whole class of threat models---such as discouraging=20
>random entities from destroying the cable due to a bungled tap, ......
That's a good argument, in peacetime anyway. OTOH in wartime, a link that=
=20
can't be tapped is _more_ likely to be destroyed.
---------------------------------------------------------------------
The Cryptography Mailing List
Unsubscribe by sending "unsubscribe cryptography" to majordomo@wasabisystems.com
