Active Networks

Rohit Khare (
Tue, 11 Feb 1997 10:53:58 -0500

Really good trip report on OSDI'96 in the current USENIX ;login: magazine.
Invited talk was Tennenhouse, Active Networks. You remember that he was
talking about smart routing, etc using IP packets with TCL in extension
headers. Mobile agents at IP level, kind of. Seems like there's lots of
ARPA money will begin flowin here.

PEP is Active Networks at the application layer. It's recursive, right?
ratehr than a packet switch/router, HTTP can be an application message
switch/router with extensibility. Perhaps some money can be found for PEP
and a generic product:

Web Exchange: can we replace the HTTP synchronous paradigm with a mesh of
peered HTTP servers, like mail/news. It would be a target that matches the
vision of Infospheres, but I don't know if Mani would like hacking 'boring
old on-the-wire technology'. This is what I want to build, so in the next
few weeks I need a clear enough picture to go to Mani (and UCI and ...) and
negotiate whether it's acceptable. Within the 'webnet', we can do caching,
routing, multicasting, middleware gateways to other comm layers,
extensibility to new message type and classes, and would handle sync AND
async messages, push AND pull distribution, unicast AND multicast, etc --
forming an 'infosphere' over *existing* protocols, not just a new
language-specific API. Basically, I think we can quickly lead in any part
of the Web research agenda, which W3C can NOT.

Also, I have a chip for Alain:

Aim at an RF-band communicator CPU. In the short term, allow for an ad-hoc
network chip. It should autoadapt to net topology and be able to
store-and-forward short messages. The key is first coming up with some sw
routing structure and encoding it in HW. The reason I think RF is critical
is that when you think about mobile munchkins, you have very rapidly
changing channel-conflict overlap graphs. It turns into a coloring problem
to choose the right radio channels to even communicate -- and then there's
a 'network-layer' routing problem. The rf link layer also needs to be
continuously adapting to changing 'cell sizes'. This new chip really is
about a drastic change in how cellular communication works: no central

This chip would certainly further my own purposes :-) but it would also be
a real advance in the wireless community (new serverless model), vlsi
(RF-band integration), an application design (drives Web

The goal is extremely cheap autonomous, distributed, communicators.
Anywhere there's a walkie-talkie, we can airdrop an agile network
infrastructure. Goal: can we establish a link b/w a tank and an F-16 in the
few seconds they are within range and exchange "webmail" of the battle map?
Can we give a hundred units to a thousand-person highschool and set up a
campus phone system? Can the system work both at high power (order
neighborhoods) and low (order people at Dodger stadium).

Part of it is just a new 'router chip', except with a new 'flood' or
'promiscuous' mode message exchange; the other is an agile rf/ir link
maintanence subsystem. Kind of like the 'ethernet' of RF needs to be

I think the cellular comm folks (CDMA/TDMA) already may have 75% of the RF
allocation problem addressed, but they're too married to fixed base
stations and master-slave relationships. Peer cellular is really cool, and
offers some collaboration potential with other departments, too.


Rohit Khare -- World Wide Web Consortium -- Technical Staff
w: 617/253-5884  --   f: 617/258-5999   --  h: 617/491-5030
NE43-344,  MIT LCS,  545 Tech Square,  Cambridge,  MA 02139