ems@amdahl.UUCP (ems) (01/07/86)
With all this talk of blowing away various news groups and
the great cost of phone calls for the backbone sites, a
question has come to mind.
How much does it cost to send an average/typical message
over the whole net?
As a 'user' of the net, I think some about the cost; but
I have no magnetude to hang onto. Just a vague feeling
that this must cost some. It would be very helpful to
me, at least, to know how much I was spending for a typical
30 line posting. If I knew that the total cost for all
sites for disk space and phone calls was, say, $24, for
some hopeless blather; I would tend not to spend it...
(If it was $1.98 though ... :-)
Does this information exist? If so, could it be advertised in
the other groups? Thanks!
--
E. Michael Smith ...!{hplabs,ihnp4,amd,nsc}!amdahl!ems
This is the obligatory disclaimer of everything.lauren@vortex.UUCP (Lauren Weinstein) (01/08/86)
Some preliminary stats I worked out once (which are only a very rough approximation, for obvious reasons) indicated that the cost of a one screen message, distributed as "netnews" to all sites, probably cost at LEAST 100's of dollars. --Lauren--
reid@glacier.ARPA (Brian Reid) (01/09/86)
Here are two independent computations of the cost of a a news message. They yield $27 per message and $32 per message, respectively (for a 1000-character message). The first computation is a forward computation based on the price of telephone calls and the connectivity of the network; the second computation is a backward computation based on Glacier's monthly phone bills and the amount of monthly traffic that we see. Forward computation: assume that... N=4000 Number of nodes on USENET C=1.5 Average redundant connectivity of a USENET node K=0.10 $US per minute for long-distance telephone time L=0.02 $US per minute for local telephone time S=80 Characters per second average throughput P=1000 characters in a news message Optimistic computation (all telephone calls local, no redundant connectivity). In this case there will be about N phone calls for N hosts; the cost of sending a message to those N hosts is N*L*P/(60*S), which is 4000*0.02*1000/(60*80) = $16.66 per 1000-character message. Pessimistic computation (all telephone calls long-distance; 50% redundant connectivity). This is N*C*K*P/(60*S), which is 4000*0.10*1000*1.5/(60*80), or $125 per 1000-character message. I think that a reasonable model--perhaps right to within a factor of 3--is that 10% of the phone calls are long distance, that there is very little local redundancy but about a factor of 1.5 long-distance redundancy (most of the backbone hosts have more than one long-distance feed path). This gives us the hybrid formula COST = 0.9*(4000*0.02*1000/(60*80)) + 0.1*4000*0.10*1000*1.5/(60*80) or COST = 14.94 + 12.50 = $27 for a 1000-character message. Backward computation: Glacier's phone bills for USENET are about $500/month and the monthly traffic is about 20 megabytes. That means that we are paying $500/20000 per 1000 characters, which is about 2 cents a message. If every node on the net pays a comparable amount, then the cost of that message is 4000*0.02 or $80/message. Maybe only 10% of the sites on the net have phone bills as large as $500/month--say the average monthly phone bill for a USENET site is only $200. That would result in a figure of $32 per 1000/character message. P.S: this message is 2500 characters as it leaves my terminal. That means it is costing about $100. I'd better shut up. -- Brian Reid decwrl!glacier!reid Stanford reid@SU-Glacier.ARPA
hes@ecsvax.UUCP (01/10/86)
I'm sure that Lauren's cost estimate would by dwarfed by an estimate of the salary cost of all the time spent by readers of that screen. --henry
lauren@vortex.UUCP (Lauren Weinstein) (01/13/86)
Golly. That's an interesting thought. What does it cost in terms of people time/salary time for the various non-technical groups that people read and post to? I'm not saying that they shouldn't-- just that the figures might be rather "alarming" in aggregrate. --Lauren--
spw2562@ritcv.UUCP (Fishhook) (01/14/86)
How about a comparison of cost to mail a message as opposed to
cost of posting an article? Maybe this would impress on our minds
how important it is to respond by mail if something is not of
interest to the entire net. And to use distributions to limit the
posting to areas of interest.
==============================================================================
Steve Wall @ Rochester Institute of Technology
Usenet: ..!rochester!ritcv!spw2562 (Fishhook) Unix 4.2 BSD
BITNET: SPW2562@RITVAXC (Snoopy) VAX/VMS 4.2rb@ccivax.UUCP (rex ballard) (02/01/86)
In article <2963@glacier.ARPA> reid@glacier.UUCP (Brian Reid) writes: >Forward computation: assume that... >N=4000 Number of nodes on USENET >C=1.5 Average redundant connectivity of a USENET node >K=0.10 $US per minute for long-distance telephone time >L=0.02 $US per minute for local telephone time >S=80 Characters per second average throughput >P=1000 characters in a news message >4000*0.02*1000/(60*80) = $16.66 per 1000-character message. >Pessimistic computation (all telephone calls long-distance; 50% redundant >connectivity). This is N*C*K*P/(60*S), which is 4000*0.10*1000*1.5/(60*80), >or $125 per 1000-character message. >I think that a reasonable model...that 10% of the phone calls are long distance >This gives us the hybrid formula > COST = 0.9*(4000*0.02*1000/(60*80)) + 0.1*4000*0.10*1000*1.5/(60*80) >or COST = 14.94 + 12.50 = $27 for a 1000-character message. >Backward computation: >That would result in a figure of $32 per 1000/character message. Two possible solutions here: Reduce Cost Per Message: 1: Crunching or compressing data might save 10-50%. COST = 4000*0.10*1000*1.5/(60*80*2) = $62.50 (worst case) 2: Trellis coded modems have an average throughput of 960 characters per second (With reliability checks). This could cut costs by a factor of ten, work on any two wire phone, and require simple RS-232 connections. COST = 4000*0.10*1000*1.5/(60*960) = $10.42 (worst case) COST = 4000*0.02*1000*1.5/(60*960) = $2.08 (best case) 3: ISDN modems, average throughput 5000 characters per second, available as leased lines, through PBX, or ISDN carriers - could cut costs to back-bone sites to 1% of original cost. Only systems with this capability should be back-bone sites. COST = 4000*0.10*1000*1.5/(60*5000) = $2 (worst case). Adding compression may or may not reduce 2 and 3 further Reduce traffic: 1: Reduce "Macro Expansion" - cross-postings, use of RCS-style 'diff -e' of previously posted articles, ability to cross reference articles. Links between articles, and tar with compression could reduce the duplication associated with 'bad habits' and stimulate interest between groups. 2: Traffic controls. Such tactics as requiring 'spell', deleting newsgroups, moderated feeds, and groups have all been suggested as ways to 'make it harder to post an article' and make the net less of a 'political forum'. Perhaps distribution menus, and routing of responses would help as well. 3: Message concentration/distribution - should articles be posted via ucbvax!allegra!siesmo!rochester!ritcv!..... if they could be posted via ucbvax!siesmo!ritcv ? Better subnet addressing would help. 4: Mailing Lists - for some reason, there are a number of 'mailing lists' which are really 'newsgroups' but get distributed in the worst possible way, via mail. Instead of making it harder to start/post newsgroups, it should be made easier. Imagine 200 identical letters going to the same machine via the longest possible path. Info-mac and info-atari or whatever they are called are just two good examples of these 'underground nets'. ARPA point to point drops are another example of potential wastes. 5: Encourage (build in to news) local networking. Local traffic is already light, but providing a full compliment of distribution to a hierarchy of local networks (Site,Company,City,State, country,world) would allow posting of information, discussion at a local level. Provide a 'long name to net name' translation directory that users could use for setting distribution. Sort of a network version of 'finger'. 6: Selective Follow-up. When a 'info-wanted' type of message come in, have a method of sending a reply to the host of the requester. If another person reading the 'info-wanted' would also like to see the responses, a 'get more info' could be sent. Each follow-up could be sent 'back up' by the machines that have that info. Computer traffic management is not that much different than automobile traffic management. Build expressways where the traffic will be heavy so the side streets won't be filled with cars. Hope this was of some use to somebody.