[e2e] New approach to diffserv...

David P. Reed dpreed at reed.com
Mon Jun 17 00:30:40 PDT 2002


At 07:24 PM 6/16/2002 +0100, Panos GEVROS wrote:
>it should be obvious that the problem lies in the user/provider boundary - 
>the
>provider has little (or no) control over how the user/client uses the 
>resource
>(network connectivity) and little information about how valuable this 
>resource
>is to the client.

When I put a letter in the mail, the post office also has no control over 
what I send, and does not know how valuable this particular letter 
is.   Why should the network provider?

It is apparent to me that this kind of thinking is contaminated by a crazy 
notion that the user/customer must serve the provider's needs.   The 
provider of transport may indeed "ask" for information, but the user is 
only required to provide a minimum of information.

What you (along with Doran and others) are proposing (and the middlebox 
layers are implementing) is an architecture based on a strong notion of 
control based in entitlement.   It is not demonstrably based in any actual 
need for the operator to know such things as whether a data packet is part 
of a video broadcast or a pornographic novel.

The "feudal system" in which the lord of the manor controls all the land, 
lets serfs farm it, and then controls the commerce they are allowed to 
engage in is not a market economy.   And its failed everywhere it has been 
tried (including the US South after 1865).   It even fails in large 
companies where the IT department tries to decide what information 
processes are allowed and what ones are not.

Building a political philosophy of control into a network architecture 
(like the middleboxes try to do) is a huge mistake.   And it is also 
unnecessary.



>since the controls cannot be put on the user machine eventually they  find
>their place in the provider's space.
>and the easiest way to deal with this is by doing something with user 
>traffic,
>system identification solutions based on the observation of external 
>behaviour
>(traffic pattern)   i.e inferring legitimate use of the resource according to
>contract or determining the value the customer derives from its connection to
>the network) are compilacated or in some cases impossible.

Why is the network provider trying to determine "legitimate use"?


>if there was a method allowing for provider control in what today is the
>client space (machine - network) then middleboxes would be redundant

Here is where your real philosophy comes out.   I suppose AOL is right in 
saying that they ought to decide whether what people send each other in 
email is acceptable or not.


>Panos




More information about the end2end-interest mailing list