The Great Exchange

"Pickett, David" writes:

1) More needs to be done to leverage locality of traffic

In the long run, why are we assuming there will be locality of
traffic?

It is true that the old PSTN has locality of traffic, but it doesn't
have flat rate pricing, or the usage patterns that the Internet has. I
argue that users are rarely more likely to be trying to download a web
page from near to their homes than from far away. If there is
locality, it is probably weak, and in the long run would only account
for a fraction of the traffic.

Perry

The moving finger of Perry E Metzger, having written:

    > "Pickett, David" writes:
    >> 1) More needs to be done to leverage locality of traffic

    > In the long run, why are we assuming there will be locality of
    > traffic?

    > It is true that the old PSTN has locality of traffic, but it doesn't
    > have flat rate pricing, or the usage patterns that the Internet has. I
    > argue that users are rarely more likely to be trying to download a web
    > page from near to their homes than from far away. If there is
    > locality, it is probably weak, and in the long run would only account
    > for a fraction of the traffic.

    > Perry

As to how much of the traffic is local, about half of the people I
know at least check email from home. Some are "weekend and evening
telecommuters", making quite extensive use of "local connectivity".

How this compares to the generic web surfing, I can't say. The CAIDA
and NLANR folks say that http traffic is the most significant portion
of today's Internet.

Perhaps someone who is actually running a local exchange can report on
how much traffic they are carrying that is now not being sent to a
MAE-equivalent? I think that actual experience and hard data will
surprise us all.

--tep

The Web is Not The Net.

Please write that 100K times on your blackboard. (PS: no, you _can't_
use expect(1l) :-).

The fact, however, that you're correct in your implication that it's
difficult to prove how much traffic would be geographically local given
the routing technologies currently available is why I was going to slip
the peering in under the rug of selling the transit -- which everyone
seems to be telling me won't work.

On another point, it's worth noting that, _currently_, all the "good"
servers are somewhere else... but this won't be the case forever.

Cheers,
-- jra

I'd guestimate that local peering and stuff accounts for as much as 5 to
15% of our traffic.

Perhaps someone who is actually running a local exchange can report on
how much traffic they are carrying that is now not being sent to a
MAE-equivalent? I think that actual experience and hard data will
surprise us all.

--tep

-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
                  Atheism is a non-prophet organization.
       I route, therefore I am.
       Alex Rubenstein, alex@nac.net, KC2BUO, ISP/C Charter Member
               Father of the Network and Head Bottle-Washer
     Net Access Corporation, 9 Mt. Pleasant Tpk., Denville, NJ 07834
Don't choose a spineless ISP! We have more backbone! http://www.nac.net
-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --

"Pickett, David" writes:
> 1) More needs to be done to leverage locality of traffic

In the long run, why are we assuming there will be locality of
traffic?

I don't know how carefully this topic has been studied, but in
the UK, where it is easy to distinguish local traffic from non-UK
traffic, there most certainly is locality of traffic.

It's easy to distinguish local traffic because nearly all UK
traffic flows to the LINX, still the UK's major peering point;
other traffic goes to international circuits. Most larger ISPs
will have one or two LINX connections and one or more
international links. It's a no-brainer to count the packets going
to each. These counts ignore packets staying within a given
network.

As far as I know, for all larger UK ISPs (those with good peering),
most traffic stays in the UK.

It is true that the old PSTN has locality of traffic, but it doesn't
have flat rate pricing, or the usage patterns that the Internet has. I
argue that users are rarely more likely to be trying to download a web
page from near to their homes than from far away.

Terms like "near" and "far away" are uncomfortably vague in this
context. However, I am sure that UK users are most likely to hit
UK Web pages first, then US Web pages, with (say) French Web pages
far behind either. And Calais is 20 miles away, whereas the USA is
3500 or so.

                                                 If there is
locality, it is probably weak, and in the long run would only account
for a fraction of the traffic.

Our experience isn't like this. The effects of locality are obvious,
but locality is not always just a function of georgraphy.

Especially not if you define "good" servers as those most frequently
accessed because you can set up a Squid cache at the local exchange point
and if every ISP connected to the local exchange point runs a cache using
yours as a parent then the "good" servers miraculously become local
servers.

The Australians have considerable experience at doing just that including
preloading their parent caches, using cheaper one-way satellite bandwidth
to load the caches (skycache.com anyone?) and setting up a national
backbone between exchange points so that the exchange point caches can all
have sibling relationships over a controlled network infrastructure. The
folks at http://www.auix.net/ can tell you more, and if you would sign up
for the NANOG meeting in Dearborn you could talk to Andrew Khoo
andrew@aussie.net and find out more.

In article <199805271744.NAA21312@jekyll.piermont.com>,

I'd guestimate that local peering and stuff accounts for as much as 5 to
15% of our traffic.

I don't see numbers quite that high with local peering, but would
guestimate that 50% or more of out nntp traffic is across a local
exchange. I'm hoping to move another 20%+ of my total traffic off onto
AADS next week, which by no means is a local exchange, but Chicago ISP's
have never been able to agree on local peering, so AADS it is for Chicago
still.

Because that has been the primary driver to date of the kind of growth
the Internet has undergone. That it has distance sensitive costs only
matters if you're trying to be a nationwide ultra-backbone. If you
just run your little local exchange, and run a couple T's to the next
couple nearby exchanges, then the fact that the loops are mileage rated
is spread out over everyone...

and the "backbone" would be much more resistant to backhoe fades...

and the "big 5" would be _really_ pissed off. Good.

Cheers,
-- jra

You seems to be ignoring the single largest cause of rtaffic, e-mail. There
is plenty of locality there.

In article <19980528100840.19859@scfn.thpl.lib.fl.us>,
[...]

> Why are you assuming that the Internet will continue to have
> non-distance-sensitive pricing, when it clearly has distance-
> sensitive costs (ultimately)?

Because that has been the primary driver to date of the kind of growth
the Internet has undergone.

Mature markets and new markets behave differently. $19.95 unlimited
dialup was a big factor in 1995 and much less so now.

That it has distance sensitive costs only
matters if you're trying to be a nationwide ultra-backbone.

Actually I was thinking about international ultrabackbones.

I don't see any theory or evidence that they are going away or being
marginalized. You will always have to deal with them. It's in their
interest to be sure that the pricing is fair. I think pricing based
on the actual destination makeup of your traffic is more fair than pricing
based on the assumption that your traffic is like everyone else's.

I could be wrong, and I'd be happy to be shown why. But I think
distance-pricing has a sound basis, even if it never materializes in
the market.

The major reason I see that it might not happen is that customers
would tend to remain fairly similar to each other in terms of their
locality; just as when customers have similar bandwidth usage,
circuits tend to be sold unmetered.

If I were an Australian ISP, I'd look very seriously at pricing
transoceanic packets different from intracontinental packets. From
what I understand the market there would be accepting; and unless
there is a flaw in the reasoning of my message yesterday, you can
collect that data cheaply with today's technology.

and the "big 5" would be _really_ pissed off. Good.

You think ultrabackbones are doomed? Why?

> Because that has been the primary driver to date of the kind of growth
> the Internet has undergone.

Mature markets and new markets behave differently. $19.95 unlimited
dialup was a big factor in 1995 and much less so now.

WHat the number is is less so, but I think the mass inxodus (opposite
of exodus, no? :slight_smile: to AOL when they wen tflat rate gives the lie to
this theory.

I don't see any theory or evidence that they are going away or being
marginalized. You will always have to deal with them. It's in their
interest to be sure that the pricing is fair. I think pricing based
on the actual destination makeup of your traffic is more fair than pricing
based on the assumption that your traffic is like everyone else's.

It may well be "more fair", but that doesn't mean it will be more
popular. People are used to a flat rate internet, at whatever level,
at least in the US, and while I'm aware that that's a horribly
Americocentric view, a) we do have the hight penetration on the planet,
by a wide margin, and b) this _is_ _NA_NOG.

It's also worth noting that the penetration of the Internet at the
retail dial up level is much higher in areas with flat rate local
service.

I could be wrong, and I'd be happy to be shown why. But I think
distance-pricing has a sound basis, even if it never materializes in
the market.

Didn't say I didn't think it was sound... but see my next reply.

> and the "big 5" would be _really_ pissed off. Good.

You think ultrabackbones are doomed? Why?

Frankly, I think that they violate some of the fundamental
architectural principles on which the Internet is based, because it is
in their commercial best interests to do so. If they connected lots of
little local exchanges, I'd be happier, but the way things are now, all
the packets end up on the shelf at one of the 4 naps...

But I've banged this drum enough this week, I'm told. :wink:

Cheers,
-- jra