Path: utzoo!mnetor!uunet!husc6!cmcl2!brl-adm!umd5!ames!pasteur!ucbvax!TWG.COM!kzm
From: kzm@TWG.COM
Newsgroups: comp.protocols.tcp-ip
Subject: Re: My two cents about charge schemes on the Arpanet
Message-ID: <8804260929.AA25210@ucbvax.Berkeley.EDU>
Date: 26 Apr 88 05:48:34 GMT
Sender: usenet@ucbvax.BERKELEY.EDU
Organization: The Internet
Lines: 36


> I know that my users will freak if we suddenly restrict them to 'free
> call' sites only.  Especially if gateways being up/down make day-to-day
> differences.  (Huh?  Why did I get billed for this telnet session to
> BNL?  It always uses Nysernet.  Sorry Charlie, that day a router was
> down at Cornell and it went via MilNet).

You raise an interesting issue : the mixing of free networks and 
charge-back networks.  In practice this is bound to occur when the
first charge-back scheme gets introduced.  Even if all networks became
charge-back after some period of time, it's hardly likely that they will 
all charge the same amount.

This points out (in my view) that charging within an internet (consisting 
of multiple separately-administered networks) cannot be done at the 
application layer (i.e. charging for individual Telnet/FTP sessions), 
but rather must be done at the IP layer.  Since the IP layer is datagram 
oriented, charging will have to be done on the basis of packets sent/received 
(but not necessarily based on packet-counts, e.g. charging could be based 
on time periods during which packets were sent).

There's also the spectre of a central WAN administration charging each of
its neighbouring administrations, which might pass on the charges to its 
users, some of whom might be other adminstrations, and etc. !!!

Given that users need to be able to specify whether (and how much ?) they
are willing to pay, it would appear that the decision to route a packet 
across a charge-back network must be made by IP routers, and therefore 
must be made based on the content of a packet's IP header.  If so, IP's
Type-of-Service could be the right place for the information to be encoded 
(e.g. "cost" to be added to the existing Delay/Throughput/Reliability, 
although one bit is probably not enough information, especially if 
"collect" were encoded here also).

Keith McCloghrie
The Wollongong Group.