🚀 go-pugleaf

In article <90Feb5.222929est.827@church.csri.toronto.edu> moraes@csri.toronto.edu (Mark Moraes) writes:
>+ a powerful CPU. (handling a full feed of news can cripple
>anything under a Sun3/180 without much effort right now)

When people say things like this, I can't help remembering that our poor
old pdp11/44, running C News, was still functioning quite well -- passing
a full feed to half a dozen other sites -- when we set it aside in favor
of a 3/180 a year and a half ago.  Volume has grown since then, yes, but
it still looks to me like disk bandwidth and good serial i/o are much more
important than CPU crunch.  The apparent importance of CPU speed is mostly,
I think, because small CPUs tend to come with slow disks and cruddy serial
ports.  (The old utzoo had Eagles on a very good controller, probably
better than the one on our Sun.  It also had Emulex DHs for terminal
multiplexors, and those make Sun's ALMs look like the crap they are.
I'm still disgusted that a 3/180 can't run a 2400-baud uucp line at full
speed -- the 44 could.)  Forget the CPU; it's the peripherals that matter
for this application.
-- 
SVR4:  every feature you ever |     Henry Spencer at U of Toronto Zoology
wanted, and plenty you didn't.| uunet!attcan!utzoo!henry henry@zoo.toronto.edu

Dream on if you think you can offer Rick 100 customers at 100 per month.

A few months ago UUNET reported about $50K/month in revenues.

Do you think Canada could account for 20% of that?  Not likely.  Face it
folks, we are a lot less unix-saturated that the USA, particularly the
area of the USA that uunet is in.

They built their business partly on the fact that they can sell you uucp
time for less than you can pay in phone bills to call somebody LD who
gives it to you for free.  They did this by buying wads of 800 number time
in bulk, and getting a fat discount.

That just isn't nearly as possible here in Canada.   If Datapac got their
act together it might be, but unless you are 100% efficient, and don't use
the datapac dial net, datapac is more expensive than a telebit call.

And we live in a country where it's cheaper to call California from Toronto
than Winnipeg.

Something is possible, but I don't know about Rick's proposal.

One idea would be to have ONet hand over operation to a private firm,
the same way that Nysernet and CAPnet did.   ONet "owns" very little,
as I understand it, perhaps some software and a router or two.  Most of
ONet's "assets" are actually things rented from Bell etc.   I understand
most of the routers are member owned, too.  (Correct me if I am wrong)

Same was true of NYSERNET.

Such a private firm could then do, at little incremental cost, what
PSI (which managaes NYSERNET) is doing -- set up uunet like services.

Actually, they're setting up more than that.  They're running white pages
and database services, and putting terminal servers in various cities.

ie. if you have an account on a NYSERNET machine, and you're in another
city with a node on that net, you can dial a local number and telnet to
your machine.   Like datapac, but part of the fixed price deal.

I don't have the time now to offer something like this on my own.  What I'm
doing now consumes plenty of my time.  But if some people want to get
together, I would be interested in joining.
-- 
Brad Templeton, ClariNet Communications Corp. -- Waterloo, Ontario 519/884-7473

Henry is right.  A Xenix/286 machine can do a decent news feed for a few
sites, if it has nothing else to do, even with B news.  With C it is
probably even more capable. 

But since you can now buy a 386/20 mhz machine for under $1K in the USA
(ok, with no disk, so 2K gets you a disk and ethernet card) -- and this machine
could probably feed 20 sites if it had nothing else to do -- it's no big deal.

For local feeds you can be quite a bit more efficient.  If a Toronto site
were to insist on calling all its customers (local call) instead of being
called by them, you can get away with lots less hardware.  In fact, I would
guess that 1 telebit and one phone line could do it.  The reason is that you
keep the load level.  One outgoing call at any given time, but it is *always*
going.  You just go in a loop calling and updating (unless a delay is desired)

For example a full USENET feed, around 7 megs, is less than one hour of
Telebit time, when compressed.  Thus a single line and modem can feed
at least 20 local sites in this manner.  (Mail is extra)

This is extreme -- you would want more than one line in case of hardware
failure, and more than one machine, but you get the point.
-- 
Brad Templeton, ClariNet Communications Corp. -- Waterloo, Ontario 519/884-7473

In article <1990Feb6.171326.15603@utzoo.uucp>
henry@utzoo.uucp (Henry Spencer) writes:

> I'm still disgusted that a 3/180 can't run a 2400-baud uucp line at full
> speed -- the 44 could.)

Gack!  That's truly appalling.  (I've never had to deal with Suns, and
wasn't aware how awful their serial I/O stuff was.)

We can handle uucp at 2400 baud on our IBM XTs!  (With Venix/86, that is.
I'm fairly certain that DOS+UUPC can go even faster, but it's undoubtedly
a fairer comparison to match the two UNIXes.)  And we all know what a pile
of junk those boxes are.  What's more, the 2400 baud is with the standard
IBM serial card, which easily meets the cruddiness characteristics of the
computer itself; you can certainly do 4800 baud uucp (under Venix) if you
equip your XT with a decent serial I/O board; possibly 9600, I'm not sure.
I presume that the 286 boxes (not quite as appallingly bad as the XTs, but
still not terribly likely to win any major awards) would do better yet.

I hope the designers at Sun are appropriately embarrassed.

Mark Bartelt                    INTERNET: mark@sickkids.toronto.edu
Hospital for Sick Children                mark@sickkids.utoronto.ca
598-6442                        UUCP: {utzoo,utgpu,lsuc}!sickkids!mark

In article <387@sickkids.UUCP> mark@sickkids.UUCP (Mark Bartelt) writes:
>> I'm still disgusted that a 3/180 can't run a 2400-baud uucp line at full
>> speed -- the 44 could.)
>
>Gack!  That's truly appalling.  (I've never had to deal with Suns, and
>wasn't aware how awful their serial I/O stuff was.)

I think it's a combination of a dumb multiplexor and a dumb driver.  Sun
believes in workstations and Ethernets and doesn't believe in timesharing
and serial lines, so the mux driver didn't get a lot of attention.  Of
course, even a workstation occasionally has to talk to someone serially,
which is why most Suns have a pair of serial ports on the CPU board...
and although the hardware behind those ports is even simpler than that
on the mux board, the *driver* has been carefully and painstakingly greased
up and *those* ports run very well at high speeds.
-- 
SVR4:  every feature you ever |     Henry Spencer at U of Toronto Zoology
wanted, and plenty you didn't.| uunet!attcan!utzoo!henry henry@zoo.toronto.edu

>I think it's a combination of a dumb multiplexor and a dumb driver.  

The real solution is to never buy ALM boards and get Annexes. They are
about the same price and do a vastly better job. Excellent product.

In article <387@sickkids.UUCP> mark@sickkids.UUCP (Mark Bartelt) writes:
|[...]
|We can handle uucp at 2400 baud on our IBM XTs!  (With Venix/86, that is.
|I'm fairly certain that DOS+UUPC can go even faster, but it's undoubtedly
|a fairer comparison to match the two UNIXes.)  And we all know what a pile
|of junk those boxes are.  What's more, the 2400 baud is with the standard
|IBM serial card, which easily meets the cruddiness characteristics of the
|computer itself; you can certainly do 4800 baud uucp (under Venix) if you
|equip your XT with a decent serial I/O board; possibly 9600, I'm not sure.
|I presume that the 286 boxes (not quite as appallingly bad as the XTs, but
|still not terribly likely to win any major awards) would do better yet.
|
|I hope the designers at Sun are appropriately embarrassed.

	Commodore Amigas have extermely fast serial
	ports - they'll handle MIDI flat out (32Kbps),
	and will reliably transfer over 1200 character
	per second, even with Amix. They appear to have
	better serial i/o capacity than the AT&T 3B1,
	which certainly is no slouch either...

Cheers,
-- 
  (__)	 Bruce Becker	Toronto, Ont.
w \@@/	 Internet: bdb@becker.UUCP, bruce@gpu.utcs.toronto.edu
 `/v/-e	 BitNet:   BECKER@HUMBER.BITNET
_/  \_	 Well I didn't want to mention it cause it's so silly, but you know how

Someone (henry?) saith:
>>I think it's a combination of a dumb multiplexor and a dumb driver.  

eric@ists.ists.ca (Eric M. Carroll) writes:
>The real solution is to never buy ALM boards and get Annexes. They are
>about the same price and do a vastly better job. Excellent product.


Agreed!  Anybody want to trade for some ALM-1's?

--dave (alas, I have six) c-b
-- 
David Collier-Brown,  | davecb@yunexus, ...!yunexus!davecb or
72 Abitibi Ave.,      | {toronto area...}lethe!dave 
Willowdale, Ontario,  | Joyce C-B:
CANADA. 416-223-8968  |    He's so smart he's dumb.