Database Home |
Subject: two hours, uucp, uunet, modems, the universe, and everything
To: puddle!f4.n494.z5.fidonet.org!ccml.rures (Mike Lawrie)
Date: Thu, 16 Nov 89 18:18:50 PST
From: Randy Bush
Cc: puddle!pdterry (Pat Terry)
I am less panicked about the bad call than may be appropriate, but then I'm
not paying the phone bill . To avert these kinds of problems, maybe we
should not run beta software. We thought that the benefits of a full-duplex
protocol would outweigh the risks, but they don't seem to for the moment.
Such is life on the freeware trail. I doubt the problem wil recur, but can
make no promises of course. We can always both go back to Binkley 2.30.
The message delays are annoying, though only mildly so. It has turned it into
a mail system as opposed to that 'next bench' feeling. Ah well.
Considering recent traffic levels, two hours is qbout one week's traffic. I
think that trying to do some things about the traffic levels may have more
payoff in the long run. Here are a few simple administrative suggestions:
Multiple folk seem to be subscribing to the same mailing lists. They
should get one subscription, and do the mailing list forking on your side
of the Pond as opposed to mine.
Similarly, we could gate each mailing list to its own newsgroup over on
this side, and send it over as a newsgroup/echomail. Oops, this would
not work with echomail as many are big messages (see below).
As you know, I do strongly support establishing a UUCP link as well. Each of
the two technologies has its strong and weak points, and it will be convenient
to be able to play them off to our advantage. We will have 'fun', one hopes
at less expense, and learn a lot there too. But there are some things of
which you may wish to be aware.
While FidoNet compresses echos and mail, current UN*X technology only
compresses news, and does not compress mail. As most of your traffic is
mail, this could be very costly.
With some research and experimentation, and advice from folk in uucp land,
Jacot and I may be able to hack our systems to compress mail. But it will
take some work and time. Although I do think it can be done, I am not
As I mentioned before, the uucp g protocol is far less efficient than
FidoNet's ZModem. Jacot and I should be sure to measure this when we
start to get serious about the ruccx1-m2xenix link.
Combine the lack of compression (50%), and a possible 2:1 loss in protocol
efficiency, and the phone gl!tch of the other week could seem quite small.
I strongly suport revitalizing our experiments with UN*X, but I just don't
want you to expect electronic nirvana. Comm has been a pig for as long as
I have been hacking. I do not expect radical improvement soon.
Having the Usenet newsgroups come over directly will allow you to
subscribe to those which have large messages, such as the source and
binary groups. They're large and therefore expensive, but I would think
the content worth it. Currently, the messages in these groups are too
large for the technique which we use to handle FidoNet echomail, and the
message splitter we are using works only for mail and not echomail.
Also note that your addresses will change from ..@f4.n494.fidonet.org to
something we will have to make up. Maybe ..@ruccx1.psg.uu.net.
How much disk does ruccx1 have? m2xenix has 225Mb, and that is quite
sufficient. I would guess that 120 or so would be sufficient for ruccx1,
unless you start subscribing to many newsgroups or keeping a long (over a
few days) backlog of news.
I wonder if the reason we have never had comments about allowing SA on the
net is that no one can see through the FidoNet screen. A recent book on
the world nets shows SA as a complete blank, saying "unknown". I have
wondered whether my going directly to uunet will cause more or less
notice. I will have to do it eventually, as oresoft or intelhf, the two
sites through which I link, will soon notice the growing volume.
On the subject of volume ... While watching the increase in traffic in recent
weeks (Tuesday was a megabyte when compressed!) I have come to realize that in
the long run we are supplying a large active country with email and news.
Thus we could view it as if I were the sole link to a medium chunk of the US.
Without some builtin mechanism to inhibit use (which I am not suggesting), we
may expect traffic to be approximately that of a major US site, which is
currently 4-6Mb compressed per day. This does not dismay me, but it is just
dwaning on me rather slowly. Maybe Vic would like to get us a T1 link .
[ aside: If Jacot starts playing with TCP/IP on PCs, I saved a complete IP
Router and a complete Bridge (Starlan, IP, ...) which came out in one of the
source newsgroups the other week. ]
> So Pat and I are looking into how this Zonegate runs, and I am alarmed by
> what I see. (As they say about apartheid, it cannot be amended, it must be
> scrapped.) Sure, much of the Fidonet complexity is of our own making, but it
> is a pretty poor system that receives N bytes, and uses 7*N bytes of
> workspace, and uses grossly inefficient DOS techniques to attempt to control
> the gateway activities.
I was not aware that the stuff on your side was so horrifying. It seems to
work quite well, and is technically correct. It really seems to do an amazing
amount for an 8088. As to space utilization, I just acquired two 300Mb NEC
ESDI drives with controllers (for friends) for USD 900 each, so I can not get
too panicked about wasting a few dozen megabytes at current street prices. I
imagine that things are just a tad more expensive over there.
I am sure that if you and Pat have the time to play with it, you will have it
sorted out quite reasonably. Pat seems to have an excellent handle on the
FidoNet stuff (as well as one or two other things:-).
[ One minor hint, DOS performance goes down the tubes as the size of an active
subdirectory gets over the size of the BUFFERS= parameter in CONFIG.SYS.
Classic thrashing. Deleting the files does not solve it, as the sectors are
not freed and the directory remains large. I presume that you know this, but
would have felt like an ass if you did not. ]
As to the modem, there are some tradeoffs. The question is whether to go PEP,
V.32, or PEP/V.32. I will explain a bit.
Usenet in the US is pretty much exclusively PEP above 2400 baud. This is
because the manufacturer, Telebit, built the UUCP g protocol directly into
their modem and offered hefty discounts to uucp sites. Hence, my newsfeed
is unquestionably PEP.
ruccx1 is V.32. This is not a bad thing as, on good lines as we seem to
have between us, V.32 seems to give 20-40% better performance than PEP, at
least for FidoNet ZModem protocol. I do not know about uucp g protocol
performance over V.32 though.
I have written to postmaster@uunet, to whom I will be moving my main mail
feed, to ask if they support V.32 and if so, does it perform well. If
they do and it does, then maybe I will have to sacrifice some newsgroups
to keep the local news feed at 2400.
If we do go V.32, then it may require some software development. You will
note that a dialer for the Telebit comes with Xenix, dialTB.c. For a
different modem, we will have to hack the Dialers file or create a dialer,
I am not sure. Please ask Jacot how he got the MultiTechs to work.
In the long run, V.32 is the better technology on which to bet, as it is
a CCITT standard as opposed to a single vendor's proprietary technique.
Unfortunately, uucp will be the last bastion of PEP to fall to V.32
because Telebit went to such lengths to cater to the uucp market's needs.
Costs are about:
PEP - USD 600 (Telebit)
V.32 - USD 625 (MultiTech; don't cry)
PEP/V.32 - USD 900 (Telebit)
So you can see why I would rather not get the combo unit, though that is
what is on 1:105/42, and works quite well. Then again, it was free as a
beta test unit from Telebit (see, being a beta tester isn't all bad:-).
I will send the form for subscription to uunet tomorrow. I imagine it will
take them a week or so to process it. This will eventually take the load of
intelhf and oresoft, maybe before they notice it .
FYI, here is the map of how I am currently connected.
uunet ------- oresoft ----------------v---- dawggon (aka 1:105/6)
`-------- intelhf --------------. | |
`--tektronix --. | | |
sun -------------- nosun!qiclab --- m2xenix --- puddle (aka 1:105/42)
rutgers --- ogccse --------'
uunet - Arlington, Virginia, the heart of the North American Internet and
uucp systems. gates to/from everywhere (eg. BITNET, Europe's EUnet)
sun - Sillycone Valley, CA, a well connected West Coast site
rutgers - New Jersey, a well connected East Coast site
others - the other sites are at local Portland companies
Currently all our mail (not news) goes/comes through either oresoft or intelhf
to/from uunet. m2xenix gets news from qiclab, which gets it from tektronix
indirectly via a number of other nodes. And remember that local calls are
free, i.e. news is free.
The plan is to connect m2xenix directly to uunet.
oresoft, intelhf, and qiclab are all PEP. puddle is PEP/V.32. dawggon is
2400. Because of the inbound Micom line at oresoft, when m2xenix or dawggon
calls oresoft it is 1200 baud, although oresoft calls out at 2400 or PEP.
As you said that you were going to work with Pat on this, I will CC him, as
what was planned to be a short note has turned into a tome with some possibly
useful data. I hope that you don't mind.