If this pipe-dream were to occur you can assume that the providers will
not be willing to provide assistance with it. They would see the number
of subscribers going down and their upstream connectivity requirements
going up as their subscriber base more efficiently utilizes bandwidth.

This being the case MPPP will not be possible, but it is also not
necessary. Out of sequence packets slowing down TCP performance is an
age-old problem on all routers doing ECMP that is solved by hashing the
src/dst addresses and ports to keep flows traveling through the same
network path avoiding out of sequence ordering. Can somebody explain why
this wouldn't work for the original posters idea? 

Please see RFC 2992 for more info....
http://www.ietf.org/rfc/rfc2992.txt?number=2992

Of course the algorithm would need to be deterministic to prevent
traffic ping-ponging between subscribers, but it would not be a hard
problem to solve. The other thing is that you wouldn't get Nx the
bandwidth for a single transfer, increased bandwidth would only be seen
with multiple transfers.

I purposely took the Manet distribution list off of my reply and would
recommend everybody else to do the same since this has nothing to do
with the scope of that list...

AF

-----Original Message-----
From: Ken Peirce [mailto:[EMAIL PROTECTED] 
Sent: Wednesday, April 02, 2003 10:10 AM
To: 'S Woodside'; [EMAIL PROTECTED]; [EMAIL PROTECTED];
[EMAIL PROTECTED]
Subject: RE: [manet] I have a dream

You're looking at combining the multiple links into one as far as TCP
can tell. This is what Multilink PPP does. Unfortunately, the problem
will be the dissimilar link speeds. Packet reordering is evil for TCP
and the slow DSL link will not only be the limiting factor but will
likely cause the bundle of links to perform like a weak DSL connection.
My $.02 Ken

-----Original Message-----
From: S Woodside [mailto:[EMAIL PROTECTED]
Sent: Tuesday, April 01, 2003 9:43 PM
To: [EMAIL PROTECTED]; [EMAIL PROTECTED]; [EMAIL PROTECTED]
Subject: [manet] I have a dream


Well, it's a fairly minor dream. What I imagine is this: I have DSL. My 
neighbour has DSL. We also both have WiFi APs that can reach each 
other, and connect at order-of-magnitude faster than the DSL links. 
What I dream is this: That we start routing internet across the link. 
When I start pulling packets off the internet, they can come to me not 
only through my own DSL but from hers as well. I think this boils down 
to an ultimate simplicity what all FN / CWNers want to achieve.

What needs to happen before I can realize my dream. I need to get an 
IPv6 address ... at least one ... for free. I need to be able to set up 
multi-homing in my home router. I need to be able to advertise my 
routing path to the internet. The internet needs to be able to handle 
at a massive scale (since every neighbour connects...). We need a /lot/ 
more IP addresses than we have now. I need to use a /real/ IP address 
because I can't route internet traffic through a NAT/site-local 
address. I need to have an ad-hoc routing algorithm that can set up 
this bridge and route across it really easily (because I'm not a 
trained network admin). Forget about BGP, AS, etc., we need a way to 
handle the massive overload of routes at local scale, at a regional 
scale, at a global scale -- something that will scale gracefully as the 
internet topology shifts from a tree to a massively connected mesh.

Please comment ... tell me what you think.

Simon

PS I'm cross-posting this to a bunch of lists because I don't know one 
it belongs on. If you think you know which is best, please tell me!

_______________________________________________
manet mailing list
[EMAIL PROTECTED]
https://www1.ietf.org/mailman/listinfo/manet
_______________________________________________
manet mailing list
[EMAIL PROTECTED]
https://www1.ietf.org/mailman/listinfo/manet
--
general wireless list, a bawug thing <http://www.bawug.org/>
[un]subscribe: http://lists.bawug.org/mailman/listinfo/wireless

Reply via email to