From: ben@rex.uokhsc.edu (Benjamin Z. Goldsteen) Subject: Re: X terminals: Suggestion for projects like SLS, MCC, etc. Date: Sat, 19 Jun 1993 08:53:18 GMT
mark@taylor.uucp (Mark A. Davis) writes:
>>>We could come up with all kinds of numbers depending on what the requirements
>>>are. I have a flexible configuration to support a wide variety of displays.
>>>This cuts costs.
>> Hmm -- 100 $1500 PC's versus 90 text terminals? I can not even
>>find a 286 anymore -- the cheapest PC's I can find are 386SX-33's. They
>>are much faster than a text terminal at text work, and they can do
>>graphics (at $400/dumb terminal, you are not doing graphics).
>Faster in what regard? They can display information faster? Not true on
>a moden text terminal. If the users don't need graphics, then why spend
>lots more money and worry about more upkeep? That was part of the point.
I believe you mentioned serial ports -- I would have to assume that
you are not connecting to your server at more than 38,400 baud (and
9,600 is more likely). The scrolling and response over a serial line
just does not compare to, say, WordPerfect running under DOS on a 286 or
386. There is a lot more work involved for one -- you have to encode
and decode the signals (I assume you are using ANSI standard terminals
like VT420's or something?. Everytime I hit a letter, the computer has
to drop to bottom of the screen and update the status line. That
happens pretty damn quick on a PC, but I do not see that happening as
quickly on a dumb terminal. You also run into those wonderful keyboard
incompatibilities (unless you restrict yourself to a pretty limited set
of keys). My point here was that you can not compare a $400 dumb
terminal to a $1500 PC -- they do not provide the same level of
capabilities
>> I am not sure what the specs are on the $875 X terminals are, but I
>>assume they are 14" B&W 70K Xstone X terminals. I assume the $1500 PC's
>>up there also have B&W monitors (good B&W monitors ones are under $200).
>They are 15" over 80K Xstones (the ones I use). Correct, the other money
>is going to software, OS, Ethernet card, etc...
>> Also, to get a server equivalent of the power in 100 PC's
>>(adjusting for multitasking) you need at least 150 SPECs.
>That is not true. In a multi-user multi-tasking environment not everyone is
>using the power of the host as the same split second. So when load is normal,
>or light, each user has access to most of the power of a fast host. During
>heavy loads the multiple CPU's kick in. During very heavy loads is the
>only time things start to degrade some. But this is rare if the host is
>carefully configured. (At least here it is)
100 PC's provide about 1500 (very distributed) SPEC's. I gave a
factor of 10 for multitasking. I thought that was fair -- I accidently
deleted the section where I said this could be achieved through multiple
processors. As for access to most of the power of a fast host...it is
wonderful to see JPEG decompression or Zip 1.9 compress on our
RS/6000... However, there are limits of how fast a machine can be --
I/O limits, lock problems, task-switching, etc -- a lot of that is OS
related, though.
Assuming that you are complete computerized and that people are
actually using the computer to do their work (unlike what most people
spend there time doing in most GUI's -- customizing their colors), what
would you consider a fairer factor? Right now, I have a 386SX-20 in
front of me emulating a VT220 (although very poorly...and the damn thing
has started dropping characters for no good reason...) -- it does not
take much of a computer to do this... On the other hand, assume our
users are running word processors, databases, spreadsheets, or various
other personal productivity applications (perhaps e-mail, DTP, graphic
design, etc).
>>You also need quite a bit of RAM.
>yes, in one place rather than scattered as much over a distributed net.
>But it is usually (overall) quite less because since everything is running
>on the host you only need to have one copy of the OS, and each running
>application.
But, you still have to have it -- I just wanted to make sure it was
included in the price. Also, PC's tend to be rather efficient with
memory (you have to be under that scourge called DOS). Last of all, it
seems like there is some duplication. My experience with X terminals is
that you need 4 or more MB left over after the server is loaded -- and
you might want some swap too so that windows do not get killed as the
terminals tries to keep itself afloat (which then contributes to network
traffic). If you have a fair amount of fonts, you could see an X
terminal with 10 MB crash, if you do not have swap, after a few
applications attack it (I have). So now, you have memory in the server
(aka host -- the thingy that runs the clients) and the terminal (the X
server).
>>For this solution to beat the cost of the PC
>>solution, we have average less than $600/X terminal or $60,000.
>Perhaps there, but not here. Like I said, the numbers can vary wildly
>depending on what your objectives are and what the environment is like.
I am not sure how this could not be true -- your X server costs
$900. The PC you quoted cost $1,500. For the hardware to come out
cheaper (which is what you quoted), the cost of the host and the
terminals must cost less then the cost of PC's. Now, you can argue that
you get different things (e.g. UPS).
>> The dumb terminal solution (dumb terminals can be X based) are MUCH
>>cheaper if your users tend not to use their computers, though (because
>>then you do not need as good a host).
>True, part of the wildly varying environment I was talking about. But
>we do make active use of perhaps half to 3/4 of all our terminals at any
>given moment during business hours.
>> My basic problem with this solution is that it uses up all its
>>bandwidth for day-to-day operations. That is, he is now utilizing 4
>>Ethernets. What happens when we want to do teleconferencing or imaging?
>>You are just not going to get 10 people doing full screen visualizations
>>over Ethernet at the same time. When you have to bring in the SGI
>>workstations, does your scheme go to poop (that is my problem with one
>>vendor solutions -- if your vendor does not provide it, you are
>>screwed)?
>Can't really address that. Don't have enough experience with using
>ethernet for non-data purposes. I suppose it could be a problem. But I
>don't find any flooding of the ethernet. Although there will be a lot of
>activity, it is very small and often. With distributed systems (especially
>diskless) the activity is less frequent, but there is a lot more info
>being transfered (files, meaning more collisions).
I am not much of a fan of diskless. Dataless is as far as I think
it should go (and that is not bad - we have a couple of dataless DG
AViiON workstations -- they feed off the main DG AViiON server -- binary
compatible so we do not have to have things like bin.aviion, bin.hp,
etc).
I am a little confused why fileserving a PC requires more of the
network then display serving an X terminal. The PC asks for nice large
blocks (though, all at once...). The X terminal will have to send over
the same information as little packets as well as all these little
packets sayings things like, "he moved the cursor." I could see a
definite win for small edits (open document, delete line 75, save,
quit), but otherwise, I do not see why the X terminal is so much more
efficient.
BTW, which X terminals have microphones or A/D converters?
Apple is coming out with some nice things where they integrate various
multimedia into your computer such that it could say do your voice-mail?
I am not quite sure what extensions there are for X to handle sound, but
I imagine X11R6 is going to have multimedia extensions?
-- Benjamin Z. Goldsteen