>Date: Wed, 21 Mar 2001 14:49:13 -0500
>From: Jerry Sobieski <[log in to unmask]>
>Subject: Re: [P2P] How much bandwidth is reasonable?
>To: [log in to unmask]
Hi Jerry,
>I agree whole heartedly with the sentiment. In fact, we had a rather spirited
>conversation at the Internet2 meeting over this topic. And fundamentally it
>came down to the cost of provisioning the the campus connection(s) to the "rest"
>of the Internet.
Yep, commodity Internet connectivity is key (and has been basically ignored).
Envision Chris Rock (the great comedian, who does that deadpan delivery of
bites from the reality sandwich of life):
"If you're going to go fast to Internet2, you're going to have to
go fast EVERYWHERE. You CAN'T just go fast to Internet2."
Tried to make this point at the Minnesota Joint Techs (see:
http://www.ncne.nlanr.net/news/workshop/2000/000515/Talks/sauver-jt05152000/ ).
Tried to make this point at the UCSD Meeting on Campus Focused Networks
(see: http://darkwing.uoregon.edu/~joe/how-to-go-fast.ppt ).
Tried to make this point during discussions of a possible role for the
proposed I2 Scavenger Service (see, for example:
http://archives.internet2.edu/guest/archives/i2ss-dt/log0102/msg00003.html ).
Tried to make this point during the peer to peer discussion at the recent
members meeting in DC (you can see my brief summary at:
http://darkwing.uoregon.edu/~joe/peer2peeronepager.pdf )
and so on, and so on, ad infinitum. I'm a broken record, what can I say. :-)
Because users do NOT know when they are using I2 or some other sort of
connectivity, you can't just provision a big pipe to I2, and a small or
medium sized pipe to the commodity Internet and expect it to work.
And why can't university's deploy large (OC3 class) commodity transit pipes?
Because they're expensive... So if buying commodity transit won't scale,
what will? Settlement free exchange-point based peering solutions.
Unless you're fortunate enough to be able to buy give-it-away-priced
commodity connectivity from one of the Cogent Communications or Yipes
of the world, Universities really need to be moving toward a settlement
free peering-based solutions... and many of them are.
The question then really becomes, "Is this anything that Internet2 can
help facilitate for the rest of 'em, assuming it wanted to do so?" and
clearly the answer is yes...
>So, in order to solve this problem, we need to recognize that Internet access is
>not free.
And yet, of course, Internet access has many characteristics that make people
behave as if it is costless.
Users get no report telling them how much they've used (unlike filling up
the gas tank, where the pump tells you how much you're using, as you
pump the gas).
Charges (if any) aren't usage sensitive -- this isn't like having dim sum
or sushi at places where you pay per plate, and you better stop when the
plates start piling up high. The current Internet model is more like Bob's
"Eat All You Want" Smorgasbord.
For many sites (and particularly for Internet2 connectivity),
overprovisioning means that usage has no direct incremental cost until
capacity reaches a step boundary (OC3 to OC12, say).
etc., etc., etc.
And who can blame users? Users think of network access the same way they
think of the campus library. They don't get charged to check out a book or
look at a journal, right? Of course, there are still costs associated with
offering campus a library (or network connectivity), those charges just
aren't usage sensitive (with rare exceptions).
>And as long as campuses provide free access, they are in fact
>subsidizing this sort of activity. Maybe that is a good thing, or at least is
>consistent with the philosophical search for knowledge we espouse.
My experience has been that most campuses do NOT provide free access, they
just don't employ usage sensitive pricing. For example, students pay
educational technology fees, departments may pay per-port installation fees,
etc. We just don't have per-port (or per-IP or per-MAC address) usage
monitoring and usage-senstive charge back.
>What would happen when the Web100 project gets really moving and starts
>automatically tuning TCP stacks for high speed links? UltraFTP...even
>legitimate users will be able to oversubscribe the link:-)
Conventionally deployed systems can do it today, using uncopywritten content,
and using already-well-known protocols (such as regular FTP). Web100's key
contribution will be in facilitating near-100Mbps *single* flows. If you
assume multiple flows, you can already get there today. :-)
>I agree with Rene. We really need to put at least part of our efforts to
>exploring ways to make bandwidth a non-issue.
Commodity internet settlement-free peering based solutions could do that for
you....
If you aren't building out a local exchange point (or peering at an
established exchange point), you're headed down the wrong road. And of
course, the only way major carriers will be willing to peer with you is
if you either individually have "a lot" of traffic for them, or if you
collectively are part of some aggregate that has "a lot" of traffic for
them. So anytime people choke back the traffic they're seeing their users
produce, they are (perversely) reducing their attractiveness as a
potential peering partner, not increasing it.
>Perhaps, such [illegitimate] over use of best effort networks will push campuses
>to explore and more agressively deploy advanced technologies to manage these
>resources such as usage based charging, QoS/CoS, line rate policy routing, etc.
I'm QoS skeptical. Edge policing issues, billing issues, and a host of
other practical problems makes deploying premium service awfully hard,
particularly on the commodity internet. And, of course, QoS/Cos can't
manufacture additional bandwidth; etc.
If the goal is to offer less-than-best effort services, there are
issues there, too, I think (largely associated with incenting people
to use that less-than-best effort service; see the discussion at:
http://archives.internet2.edu/guest/archives/i2ss-dt/log0102/msg00019.html)
I'm also of the belief that billing is a huge pain with only marginal
returns for most users (that's why we don't track and charge back water
usage at most campuses);
The only thing that does scale is to deploy settlement-free peering based
commodity connectivity.
Regards,
Joe
|