olson@lasspvax.UUCP (Todd Olson) (11/03/85)
[]
I would like to make a suggestion for restructuring the net in such a way
that everyone pays their FAIR SHARE of the PHONE COST. Let me warn you
first that I do not know how the net currently works as my thesis keeps getting
in the way of studing the matter, hence my solution might be the current
system. (Please: no flames, I'm merely a concerned citizen trying trying
to keep a 'public good' going.) Also if the problem is more the demand
the net is putting on a machine in cpu time spent serving the news then
in money paid to the Phone Company, then my 'solution' isn't.
|-->-- 1
news --->--- A --->---|-->-- 2
|-->-- 3
|-->-- 4
The current distribution method, if my inference is correct is that
after the news has arrived at A, A calls up 1,2,3,4 in turn and passes
the news on to them each in turn. Probably, at the same time A also
collects the any new news from 1,2,3,4. Thus A pays for all of the
communication between it and 1,2,3,4.
IN SHORT
I suggest this should be turned around. Machines 1,2,3,4 should each
call A and ask for any new news.
IN DETAIL
Several problems immediately come to mind, one being how do we keep
1,2,3,4 from calling A at the same time and another being how should A get
new news from 1,2,3,4. Here is what I imagine should happen...
1) For every pair of machines that exchange news directly, (eg A,1) one
is designated the up-stream site (A) and one the down-stream site (1)
2) At some (more or less) prearranged time A calls 1 and asks for any
new news that 1 has to pass along. 1 sends the news to A. The
connection is then terminated.
3) Immediately 1 calls A back and asks if A has any news for it. Now
A sends news to 1. The call is terminated.
4) The news exchange is now done.
DISCUSSION
The main point to note is that if every machine gets every piece of news
then EVERY machine PAYS the SAME in phone bills because each machine pays
only for what it gets, not for what it feeds to others.
One might object that it this takes TWO phone calls to exchange the news
where one should suffice. But TWO phone calls is the only way to make
each machine pay for their share of the news.
The reason for having one site (A) always initiate the exchange is so
that it doesn't get called simultaneously by 1,2,3,4.
Note that if A decides that it will not deal in some group, say *.mac,
and 1 wants this group, then 1 can still get most of its news from A
and then go elsewhere (possibly long distance) to get *.mac.
This scheme might make it easier for some of the backbone sites to
off load some of the serving because it would be less costly for other
sites to become servers.
PROBLEMS
1) The main backbone sites will still spend a lot of cpu time
communicating with other machines. They will still need banks of
phones to handle the traffic. (However they won't be paying for
all that traffic. Only for what they get, which is the same as
every one else)
2) What if in step (3) above, 1 does not call back. Does A hang, not
dealing with 2,3,or4? I suggest some sort of time out mechanism.
3) This scheme will encourage Fragmentation. That is it will be possible
for some site to draw it's news from several sites, for speed
or cost reasons. This will complicate book keeping. Maybe the
first bit of communication should be, "I want news from you, the
last thing I saw was ... Do you have anything more up to date
than this?"
4) Sites that feed news long distance will have higher phone bills.
(I suppose we could put up with only local calls and long transit
times if the net is dense enough in real space. (-:) Maybe then
we could get the biologist to study it for us to help us keep it
alive (-:))
5) Suppose one site generates a lot of 'junk'. They never pay for
what they generate, but everone else pays to get it. Well, I
don't see a way around this. All I can say is that at least it
is better that each site pay for receiving junk mail rather than
one backbone site (that merely transmits the junk) paying for
everyone to get the junk.
I'm sure that there are other technical problems.
Will this distribute costs more equitably?
What do the rest of you think?
Is there anyone up to creating this new beast? (Unfortunately(?), I have
a physics thesis to produce, otherwise I'm just crazy enough to try create
this on my own, for personal amusment and education.)
SUMMARY
Structure the net so that people pay for what they get rather than what
they give.
--
Todd Olson
ARPA: olson@lasspvax -- or -- olson%lasspvax.tn.cornell.edu@cu-arpa
UUCP: {ihnp4,allegra,...}!cornell!lasspvax!olson
US Mail: Dept Physics, Clark Hall, Cornell University,
Ithaca, New York 14853-2501henry@utzoo.UUCP (Henry Spencer) (11/05/85)
> The current distribution method, if my inference is correct is that > after the news has arrived at A, A calls up 1,2,3,4 in turn and passes > the news on to them each in turn. Probably, at the same time A also > collects the any new news from 1,2,3,4. Thus A pays for all of the > communication between it and 1,2,3,4... > I suggest this should be turned around. Machines 1,2,3,4 should each > call A and ask for any new news... In our particular situation, actually, this wouldn't make much difference. Machines 1-4 are all local calls. The problem is getting the news here in the first place. There are more general problems with schemes like this. The people with the worst phone bills are the backbone sites, for whom the situation isn't as simple as presented above. Backbone sites generally have redundant links to other backbone sites; those links are major factors in the phone bill. Unfortunately, this part of the bill cannot be straightforwardly charged to a single site -- it's just part of keeping the network running. Another problem is that the phone bills long ago became serious enough that most sites are reluctant to accept asymmetric Long Distance links: "if we're going to call you, you have to call us too, so we don't pay the whole bill". When it's a question of A getting the news from B, this can often be avoided by B simply refusing to budge. When the relationship is less one-sided, this isn't a viable approach. > The main point to note is that if every machine gets every piece of news > then EVERY machine PAYS the SAME in phone bills because each machine pays > only for what it gets, not for what it feeds to others. Unfortunately, phone bills are a function of distance as well as volume. It is not at all uncommon to have one site in an area that does the bulk of the long-haul transmission, with local redistribution fanning out from there. -- Henry Spencer @ U of Toronto Zoology {allegra,ihnp4,linus,decvax}!utzoo!henry
ron@brl-sem.ARPA (Ron Natalie <ron>) (11/06/85)
I'm not going to include any of this discussion because it was long
and incorrect. The author makes several invalid assumptions.
1. Netnews flows in a true tree like fashion.
Wrong, net news can and sometimes is richly connected.
2. The major expense of backbone sites is distribution to local
sites.
No, not quite. Many people poll the backbone sites and
other local sites for news. Remembering the original UUCP
paper, you can divide the net in to three classes of hosts:
Rich hosts (with autodialers), Cheap hosts (incoming modems
only), and Paranoid hosts (no modems). Handling news on
any host is a major load regardless of which way the charging
on the phone lines go. In addition, the backbones are still
paying for the high expense backbone node to backbone node calls.
-Ronglenn@ll-xn.ARPA (Glenn Adams) (11/10/85)
What would seem desirable is a means for communicating between backbone hosts other than by using phone service. For example, a few sites inter- connected via the ARPANet could greatly reduce this expense. Our site currently uses TCP/IP for incoming and outgoing news using the DOD Internet as the medium. A few well placed hosts homed to both the Internet and USENET could go a long way in reducing some of the major long distance tolls between backbone sites. -- Glenn Adams MIT Lincoln Laboratory ARPA: glenn@LL-XN.ARPA CSNET: glenn%ll-xn.arpa@csnet-relay UUCP: ...!seismo!ll-xn!glenn ...!ihnp4!houem!ll-xn!glenn
cc-06@ucbcory.BERKELEY.EDU (Ilya Goldberg) (11/10/85)
In article <239@ll-xn.ARPA> glenn@ll-xn.ARPA (Glenn Adams) writes: >What would seem desirable is a means for communicating between backbone >hosts other than by using phone service. For example, a few sites inter- >connected via the ARPANet could greatly reduce this expense. Our site >currently uses TCP/IP for incoming and outgoing news using the DOD Internet >as the medium. A few well placed hosts homed to both the Internet and >USENET could go a long way in reducing some of the major long distance >tolls between backbone sites. >-- > >Glenn Adams >MIT Lincoln Laboratory The DOD people just might have something to say about that - after all, ARPANet is not free either. Maybe some sort of a priority scheme would be in order. That way the DOD would still get all of their traffic thru in the same amount of time while any 'unused connect time' could be utilized for news. Same goes for backbone sites that can't afford the cputime for news, why not just run the news software at a lower priority or only at night? Ilya (..!ucbvax!ilya)
earle@smeagol.UUCP (Greg_Earle) (11/12/85)
My Swiftian Modest Proposal for Saving The Net : Make sure all the backbone sites in the net are either : (1) AT&T sites; -or- (2) Sites at US government organizations. This will help with at least some of the 'phone bill costs too much' whining we've been seeing. Reasons: 1) "We don't care. We don't have to. We're the Phone Company!" :. => "I don't care (about their phone bill). I don't have to. YOU'RE the Phone Company!" 2) *Everyone* gets to help pay the government's phone bills (thank you, John Q. Taxpayer), so no whining about unfair phone bills. *Everybody* gets to help pay! How democratic can you get? Hell, I'd certainly much rather see my tax $$ going to support net traffic than building Nuclear Bombs anyday. (% mail jedgarjr@FBI.ARPA << "You see that? Put him on the hit list!" ) I knew you'd love it. <;=] ------------------------------------------------------------------------ Can you say "divestiture"? Sure. I knew you could ... Greg Earle JPL ...!sdcrdcf!smeagol!lorien!earle ...!sun!tsunami!smeagol!lorien!earle
lauren@vortex.UUCP (Lauren Weinstein) (11/15/85)
DoD rules explicitly prohibit the sending of certain sorts of material on ARPANET, and also specify what may be appropriately sent. Much (possibly most) Usenet netnews traffic would not be permitted and would put the sites passing such data at considerable risk. To put it another way, it isn't worth risking an ARPANET connection for the sake of the 90% junk content of Usenet. Even if there WERE no content issues, using ARPANET to avoid paying phone charges would go directly against the rules that say ARPANET must not "compete" with commercial telecom services. --Lauren--
sob@neuro1.UUCP (Stan Barber) (11/18/85)
Lauren, I (and I think the rest of the non-ARPA sites) would love to
see a copy of these rules. Can you point to the appropriate RFC or
whatever that we can get from SRI-NIC? Or, better yet, can you
cite (i.e. quote) these rules right here in net.news (yes, this
is net.news.adm, but follow-ups are pointed at net.news).
This information isn't restricted, is it?
Or the arguement of appropriate use of ARPANET vs material on usenet,
I must concur that many of the "flame-oriented" news groups are
probably inappropriate. However, I do notice that most (if not
all) the mod groups originate in ARPA-land. Given that, could the
ARPANET be used to transport that group from backbone to backbone?
I also note in the ARPANET list of lists that there are recreational
oriented groups like SF-Lovers, RAILROAD, CUBE-LOVERS, and so on.
Even a few OPINION groups are to be found: ARMS-D, POLI-SCI
and so on. I think your estimate of 90% might have been a bit high, don't you?
In fact, it is not clear to me that some sites do not ALREADY employ
this method of distributing news via TCP-UUCP.
--
Stan uucp:{ihnp4!shell,rice}!neuro1!sob Opinions expressed
Olan ARPA:sob@rice.arpa here are ONLY mine &
Barber CIS:71565,623 BBS:(713)660-9262 noone else's.ron@brl-sem.ARPA (Ron Natalie <ron>) (11/18/85)
> Lauren, I (and I think the rest of the non-ARPA sites) would love to > see a copy of these rules. Can you point to the appropriate RFC or > whatever that we can get from SRI-NIC? Or, better yet, can you > cite (i.e. quote) these rules right here in net.news (yes, this > is net.news.adm, but follow-ups are pointed at net.news). > This information isn't restricted, is it? > There is a statement about the official use of the ARPANET that is very broad (fortunately) which allows certain of the not quite official content traffic that exists on ARPANET. It's in my manual in my office so I can't quote it, but it really doesn't quite make any difference. ARPANET traffic is restricted to official government traffic and traffic in support of the ARPANET charter. Since ARPANET is an experiment in networking, it could mean almost anything. Many moderators prefer to take a conservative view on what is allowed because they don't wish to be bounced off the net (or as is more likely to happen, incur the rath of their host sponsor and get kicked off the machine being used for the mailing list). > Or the arguement of appropriate use of ARPANET vs material on usenet, > I must concur that many of the "flame-oriented" news groups are > probably inappropriate. However, I do notice that most (if not > all) the mod groups originate in ARPA-land. Given that, could the > ARPANET be used to transport that group from backbone to backbone? Not all mod groups are ARPA-originated, but it is pretty much the case that all ARPA-originated lists are MOD lists. This is because of inadequacies in the old fa.* distribution lists. You can't use ARPANET as a common carrier to transfer between two sites on the net just to circumvent the phone charges. It would be possible to provide multiple insertion points into the USENET for traffic originating in ARPA, but this requires a little more cooperation between the ARPA sites. Erik Fair is actively working to accomplish this though. > In fact, it is not clear to me that some sites do not ALREADY employ > this method of distributing news via TCP-UUCP. BRL-TGR, BRL-SEM, and NLM-MCS are USENET machines that have no UUCP on them at all. We obtain the traffic using some shell scripts and the Berkeley RSH utility. We can do this because it has been stated that we should use DDN (thats what they call MILNET and it's connected networks, commonly called ARPANET) rather than leased lines or even dial up modems. Given two MILNET sites who wish to play backbone, they could get away with using the MILNET as the transport medium. The hard part is convincing a government site that it is in their best interest to be a part of the backbone. -Ron