leichterj@rani.DEC (12/11/84)
> Newsgroups: net.unix-wizards > Path: decwrl!decvax!wivax!cadmus!harvard!seismo!brl-tgr > !tgr!root%bostonu.csnet@csnet-relay.arpa > Subject: Re: What irks me about Unix mail > Posted: Fri Dec 7 20:11:58 1984 > If you mean that VAX MAIL under VMS using DECNET > is more reliable in its routing it is because it > encompasses a trivial network, what do you have? > 5 or 10 VMS/VAXes on a single simple spine or point-point? > Anyone can solve networking as long as they limit > the problem enough which is largely what decnet does. > > When you have hundreds, maybe thousands of nodes with > no centralized administrative control (as you would > need with decnet)... > > ... > -Barry Shein, Boston University Congratulations. You have just described DEC's internal DECnet network, the EASYNET. Last time I looked, it had something like 3,700 active nodes on it, ranging in size from small 11's running RT-11 to large VAXes to TOPS-10 and TOPS-20 systems. The EASYNET started out as the "Engineering Net", linking various development machines. These days, just about every machine at DEC is on the net, or will be soon - growth is still rapid. As you can see from this message, the EASYNET has links to UUCP; it also connects to ARPANET and CSNET. Those 3,700 nodes, by the way, are pretty well distributed. Earlier today, I received a note from someone on a machine in Tokyo about 68000 C compilers. There are a lot of EASYNET machines in Europe. Other places I've had reason to have contact with are Israel and Tokyo. Unlike UUCP, EASYNET is: (a) a full routing network. Except in some corners of the net where for various reasons - ranging from old machines to security - restrictions are placed, I need know only the name of the node I want to reach, not a path to it. EASYNET knows that. (It's pretty good at finding those paths. At one point, traffic between Maynard, Mass. and Nashua, Hew Hampshire was noticed to be slow. It turned out that the direct lines were down - so packets were going through England!) (b) more than a mail network. First off, it's at least two mail net- works. DEC provides VAX MAIL for free with VMS. VMS MAIL is admittedly a rather simple-minded mail system, though vastly improved in Version 4 of VMS. DEC also sells some more sophisticated mailers which use a product called Message Router, which provides services like queing and guaranteed delivery. Beyond that, EASYNET provides things like remote file access - transparently on VMS - and various distributed products, such as VAX VTX, a Videotext product. By now, you are probably wondering about the army of people needed to run all this. Well, central administration of the EASYNET is done by about 5 people. Most of what they do is concerned with getting and maintaining various back- bone lines, helping new systems get attached to the EASYNET, and tracking down and correcting various bugs and problems that local personal can't resolve. They also maintain a node name/number registry - something you obviously need if you are going to have automatic routing. -- Jerry
thomas@utah-gr.UUCP (Spencer W. Thomas) (12/12/84)
Yeah, I saw the DEC Engineering net in operation this summer. My friend
couldn't reply to a mail message because the machine she was replying to
was down at the time! That's worse than telephone tag (well, almost).
=S
--
=Spencer
({ihnp4,decvax}!utah-cs!thomas, thomas@utah-cs.ARPA)
<<< Silly quote of the week >>>guy@rlgvax.UUCP (Guy Harris) (12/13/84)
> > Discussion about DECNET being trivial networks, like 5-10 VMS VAXes > > on a spine or point to point, as opposed to the UUCP net having > > hundreds or thousands of nodes with no centralized administrative > > control... > Discussion of DEC's internal DECnet network, EASYNET, with ~3700 active > nodes and central administration done by 5 people... This whole thing was triggered by somebody bitching that UNIX mail was a pain because it doesn't notify you immediately that a message can't be delivered, as opposed to VMS mail. Well, EASYNET may be a nice net, and the reply did point out that one can't dismiss the difference between the two mail systems that easily, *BUT* it's still not relevant. You can't really compare EASYNET to the UUCP net. If you want to compare EASYNET to a net containing UNIX machines, try the ARPANET. DECNET is, as was pointed out, a "full routing network"; this is because there is probably no need for store-and-forward routing of mail messages, as there is with the UUCP net. If one were sending all mail over the ARPANET, one could probably get the same sort of "that site doesn't exist or can't be reached" messages immediately. If the VMS mail system started using UUCP heavily, it would have the exact same problems UNIX has. If one had a centrally-administrated network like EASYNET of UNIX systems, which permitted any machine on the net to establish a connection to any other machine using the network and transport layers of the network's protocols, it could have the same advantages as the VMS mail system. In other words, the difference between UNIX mail and VMS mail in this regard has nothing to do with UNIX or VMS - it has to do with the fact that UNIX mail permits messages to be transported using UUCP, which may have reliability problems out the wazoo, but which *does* permit you to send messages with no more investment other than in a modem and in phone charges to the UUCP sites you route mail through. Guy Harris {seismo,ihnp4,allegra}!rlgvax!guy
kupfer@ucbvax.ARPA (Mike Kupfer) (12/15/84)
> > > Discussion about DECNET being trivial networks, like 5-10 VMS VAXes > > > on a spine or point to point, as opposed to the UUCP net having > > > hundreds or thousands of nodes with no centralized administrative > > > control... > > > Discussion of DEC's internal DECnet network, EASYNET, with ~3700 active > > nodes and central administration done by 5 people... > > ... > > You can't really compare EASYNET to the UUCP net. > If you want to compare EASYNET to a net containing UNIX machines, try the > ARPANET. DECNET is, as was pointed out, a "full routing network"; this is > because there is probably no need for store-and-forward routing of mail > messages, as there is with the UUCP net. If one were sending all mail > over the ARPANET, one could probably get the same sort of "that site doesn't > exist or can't be reached" messages immediately. You don't get immediate "no such user" or "this site is down" messages when you send mail over the DARPA Internet. What typically happens is you send the bogus letter, and then you get return mail telling (sometimes cryptically) what the problem was. In the case of "no such user," the return can be as quick as a few minutes. The "host down" message typically takes longer (I've seen 3 days) because of a long timeout. Either way the return mail has the text of the message you sent, so all is not lost. Which brings me to my point. As Guy said, the problem is in using a store-and-forward net. You can argue that a virtual circuit system gives you the advantages that VMS mail gives you. Or you can argue that the mail system should queue the message and deliver it when possible. It seems silly to me to get what amounts to a busy signal just because some gateway happens to be down for P.M. just at the time I send my letter. But it also seems that uucp mail could learn a few tricks from the Internet system. -- Mike Kupfer kupfer@Berkeley ...!ucbvax!kupfer When the going gets tough, it's time for tea.
honey@down.FUN (12/16/84)
/***** down:net.unix-wizar / ucbvax!kupfer / 1:08 am Dec 16, 1984*/ ... But it also seems that uucp mail could learn a few tricks from the Internet system. /* ---------- */ uucp has learned a few tricks, not from the internet, but from common sense. the latest version of uucp notifies the issuer (*not* previoushost!uucp) of problems. if a message is stuck in the spool for a day, a warning is sent back to the issuer. after a week, the message is removed, with another notification. in both cases, the message is included in the notification. is this how the internet does it? peter
ron%BRL-TGR@tgr.UUCP (12/17/84)
Sorry, there is nothing inate to ARPA mail that keeps you from getting "no such user" or "this site is down" messages. The specification only deals with two machines talking to each other once the connection is made. Most systems will tell you that a user doesn't exist during the SMTP dialog. It is everyones implementation that deals with when retries are done and what the nature of the error message returned is. On our system, you can either send your letter into the mail queues and blindly assume it will either get there, or you'll get notified eventually, or you can sit there and watch it try to be delivered to the remote machine immediately. -Ron