RE: WG next steps

["Michel Py" <michel@arneill-py.sacramento.ca.us>]

Andrew / Masataka / Rob

>> Michel Py wrote:
>> Nobody cares about this if the free transit you
>> provide represents 0.1% of your total traffic
>> (not to mention the fact that your competitors
>> would likely give you back 0.05% anyway for the
>> very same reason.

> Andrew Partan wrote:
> It means that if there is one provider that
> connects to A and B, then no-one else in A or B
> needs to buy transit between A & B as that one
> provider will have to provide it.  That will not
> fly.
> It means that some small provider only in area A
> does not need to buy transit at all.
> I, as a large ISP, will only give transit if at
> least one end pays me.  Who is going to pay me?

[note that for clarification purposes what Iljitsch and I are talking
about is not the same thing at all]

I think that you have not grasped the concept. The model we are talking
about is that topology requests (traffic would be equivalent to the
initial SYN of a TCP connection for the session to a multihomed site,
twice) is what goes over geographical addresses and is what you *might*
have to provide the free traffic for. The bulk of the traffic is carried
over PA, which means that the small ISP would have to buy transit to
carry it, and this part is not free.

So what we are talking about potentially is that every ISP that services
a geographic zone would have to *potentially* carry the first SYN packet
for their competitor's traffic.

> Masataka Ohta wrote:
> Then, how can you say it 0.05%?

Actually, I would be happy to have some real figures about it.
Typical (if such a thing exists ;-) Internet use, what percentage of the
bandwidth does the first SYN of the first HTTP/TCP connection to the
site represents? I have the exact non-numeric answer to this: peanuts.


> You talked about bandwidth, not identifier/locator mapping.

Precisely. The identifier/locator _mapping process_ uses geographical
addresses, the actual traffic uses PA addresses.

> If you are saying that the geographical addresses can
> not be used in real packets, they are not locators but
> identifiers.

Correct. The geographic address is the identifier, there can be multiple
PA addresses that are locators.

> Instead, if you are saying that the geographical
> addresses can be used without such restriciton as
> bandwidth limitation for free riders, they will be
> used 100% of the time.

I agree with the "will", but this is not what I am saying.

> Robert J. Rockell wrote:
> I care very much.  CapEx with no associated revenue
> (even if you can make the arguement that it all comes
> out 'in the wash') is not going to fly with the
> companies left in business today...

Rob, I'm a little surprised about you posting this. We are talking about
MHAP, which you know to some extent, and I have never heard you scream
about this before.

Again, what is the issue here: you *might* *potentially* provide free
transit for the first SYN of Joe Blow surfing a web site that is hosted
by one of your competitors. Is this kind of bandwidth on your radar
screen?

Michel.