My argument was that in addition to the peering settlement charges,
each network would charge their customers for the traffic they generate.
So, if you ask your web hosting customers to add traffic, and actually
pay them for that, then my customers that request content from yours
get a bigger bill than they used to. My customers stop going to your
customers site. (I'll provide them with an itemized bill that lets them
see what cost so much and why.)
This is nice. Do you have any idea what kind of backend software would you
need to be able to process that much data in real-time?
Netflow seems to scale up to DS-3 levels with no problems, and, in fact, I
haven't seen too
many problems with it on OC-3's either... Now, the UDP export to the receiver
some issues. But with 400MHZ processors becoming common-place, I don't think
difficult to do.
I'm not saying the technology is there today, but is there any doubt it won't
HA. Very funny. You really intend to send out billing notices to your
clients that detail each web transaction they did? So you intend to send
you ISP customers the equivalent of an inverted access_log file which they
will in turn send to each dialup customer? For a single T1 this data runs
in the range of 1 GB per day. The cost of handling this billing
information is non zero and the cost of reviewing it is non zero.
I'm not saying it needs to be an inverted access-log. A simple destination
and byte count should do the trick. And of course its non-zero. Its the
same with the telco's and their settlements. It costs money to do accounting,
but if you are now charging for usage, I don't think you'll lose money.
And if you can't account on each destination and byte count, account based
on packets sent in a 10 minute interval, or something that is scaleable
today. (Yeah, you lose the destination that is sending the majority of
the traffic, but I don't think that would be too difficult for the customer
to figure out.)
But, if the majority of nanoger's feel that its impossible to do accounting
on traffic flows per customer, then I'll withdraw my argument that settlement
based peering could work when you charge your customers based on usage.