[45689] in North American Network Operators' Group
Re: Reducing Usenet Bandwidth
daemon@ATHENA.MIT.EDU (David Schwartz)
Tue Feb 12 11:49:05 2002
From: David Schwartz <davids@webmaster.com>
To: <stuart@tech.org>
Cc: <nanog@merit.edu>
Date: Tue, 12 Feb 2002 08:47:49 -0800
In-Reply-To: <200202120959.g1C9xEQ16473@hi.tech.org>
Mime-Version: 1.0
Content-Type: text/plain; charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
Message-ID: <20020212164750.AAA14603@shell.webmaster.com@whenever>
Errors-To: owner-nanog-outgoing@merit.edu
On Tue, 12 Feb 2002 01:59:14 -0800, Stephen Stuart wrote:
>>Think about it. I post a reply to a question in a newgroup. The=
more
>>intelligent and interesting it is, and the more my reputation=
makes people
>>want to read my interesting comments, the more I pay. Does that=
make any
>>sense?
>You're stuck thinking about users again. This is between sites,=
as I
>thought I explained previously. This is about my spool having=
less
>trash.
=09I'm talking about between sites. Why do you think there's a=
difference?
Ultimately, supply and demand and the inexorable laws of=
economics will cause
the cost of a product or service to track the cost to provide=
that product or
service.
>If your post is not a pirated copy of Word, and if you, as a=
user, can
>be intelligent and interesting and enhance your reputation in=
less
>than, say, 20 KB per post (you seem to be doing fine in less=
than 2KB,
>so no worries there), then I don't think the wonderful world of=
USENET
>should change for you. Not one bit.
=09I think it's wishful thinking to expect a change in the cost=
model not to
effect the change in the cost of a service. Every example I know=
of points
the other way. When ARIN started charging for IP registration,=
for example,
ISPs started charging their customers for IP space. The net=
effect was a
reduction in the use of IP space that didn't pay for itself. The=
same would
happen to USENET.
>On the other hand, I want the site that accepts postings from=
you to
>incur higher costs if you or your site-mates inject pirated=
copies of
>Word that take up space in my spool and eat up my bandwidth when=
*my*
>site's set of users and downstream feeds have no interest in=
that
>(apparently we lack "human nature").
=09So you want to charge for useless content without increasing the=
costs for
useful content. You imagine that an automated system that=
approximates worth
based on short size will meet this goal. But the obvious=
collateral effects
are that:
=091) Worthless or harmful content that is nevertheless short will=
be
legitimized. After all, the creator is paying for it.
=092) Useless and beneficial content that is nevertheless long will=
be
penalized. Why should I pay to educate you?
>On a private thread that cc'd you, I said this:
>If my server pre-fetched only the articles in the groups that=
its
>users were known to read, on the whole it would have much fewer=
large
>binaries than visit its spool every day. Behold, less trash.
=09And if you are talking about arrangements within a single=
administrative
domain, this won't change the price/cost model but will still=
reduce trash.
So you don't have to go to 'sender pays' to get what you want.
>The idea is to make the cost of injecting trash high.
=09Fine, when you invent an automated system that can sort trash=
from treasure,
you can do this. Otherwise, you'll increase the cost of injecting=
treasure as
well.
>If your site
>doesn't tend to inject lots of things that other sites tend not=
to
>want to carry in their spool (pirated copies of Word, MP3s, and
>pornography are the current set of examples), you - as a site -=
would
>be rewarded by not having many unicast hits.
=09You are now in defiance of reality, I think. I don't have hard=
data, but I'm
going to bet that more than 50% of USENET users are in it for the=
binaries.
Do you have statistics to show that your view of what is "trash"=
is anything
more than your view?
>Note that I consider unicast transfers of articles to take=
place
>between spools; when a reader asks for an article that is only
>resident as a pointer, the reader's spool would go get it and,=
in
>theory, cache it for the next reader who happened to want it.=
Probably
>some unwanted copies of Word would end up in spools this way,=
but at
>least it would be as a result of some user asking and not just=
because
>ten other sites decides my spool should have as many copies of=
Word as
>possible. Number of unicast transfers does not equal number of=
readers
>of an article; hopefully you'll see some increasing distance=
between
>what I'm saying and the belief that it would somehow change=
USENET for
>you.
=09Again, I have no reservation about this being done between=
specifically
consenting sites as a way of providing a more efficient feed=
between them. My
reservation consists of expecting A to provide unicast news=
services to C
when they have no agreement except through an intermediary B. You=
can't say,
"if you agree to feed news to me in this new highly efficient=
manner, you
must agree to provide unicast news service at your expense to=
anyone who
wants it." This would, as I see it, affect the USENET cost model=
so
dramatically as to risk making the content situation even worse.
=09Yes, it might make things better for porn and warez. But I worry=
about what
it would do to the good content that's still there. You can't=
think that
sender pays won't reduce the amount of sending that benefits only=
the
recipient.
>There might be some benefit from an academic perspective in an
>implemention that would allow a news admin to set a knob so=
that
>articles below a certain size got flooded as currently happens,=
while
>articles above got "pointerized" (headers plus overview=
records,
>perhaps) and thus only fetched if actually desired by a=
downstream
>site.
=09I agree. that could be a good idea between sites that wish to=
adopt such an
arrangement, especially between major hub servers and smaller=
leaf servers or
network. And perhaps if a few large news sites agreed only to=
accept that
type of arrangement, USENET could gradually get to that point.
>The notion is not to replace USENET with the web publishing=
model (we
>already have the web for that, and "web forums" and their ilk=
haven't
>exactly rendered USENET obsolete); if anything it's to augment=
it with
>some capabilities from that model that are there for site admins=
to
>use if they choose to do so. If that turns into costs that site=
admins
>want to recoup in the form of charges for their users, that's=
left as
>an exercise for the implementors (remembering my opinion stated=
above
>that I don't think USENET should change for you).
=09It may be instructive to think about *why* web forums can't=
replace USENET.
If you don't know, it would be a very bad idea to adopt their=
concepts. One
advantage USENET has is a more global and consistent namespace,=
and that it
would still have. But another is that distribution is=
inexpensive, even from
low-bandwidth locations, and changing that could really change=
things.
=09That said, I do realize that USENET has a lot of utter garbage=
and I'd love
a good way to reduce that.
=09DS
