Laser-Lovers-Request@WASHINGTON.ARPA (Moderator) (01/12/86)
Info-Postscript for Laser Lovers Digest Saturday, January 11, 1986 1:31PM
Volume 1, Issue 3
Today's Topics:
Dataproducts LNZ-2660?
Re: Dataproducts LNZ-2660?
RE:Dataproducts LNZ-2660?
Re: RE:Dataproducts LNZ-2660?
RE: Dataproducts printer
Printer speeds
Re: Printer speeds
Re: Printer speeds
Re: Printer speeds
Re:Printer speeds
----------------------------------------------------------------------
Date: Thu 21 Nov 85 22:04:30-EST
From: Bill Schilit <BILL@CS.COLUMBIA.EDU>
Subject: Dataproducts LNZ-2660?
Has anyone seen, or does anyone have an LNZ-2660? This is a
PostScript printer made by Dataproducts, and mentioned a while back on
this list.
Does anyone know if the 10MHZ clock cycle of the LNZ 68000 vs. the
LaserWriter's 12MHZ cycle makes a difference? Or the effect of 2MB
RAM on the LNZ vs the 1.5MB on the LaserWriter?
A number of companies are OEM'ing the Dataproducts machine under a
different name -- like Apollo computer's Domain/Laser-26 -- has anyone
seen any of these?
At $20,900 the LNZ-2660 is about equal to 4.5 LaserWriters. It is
only 3 times as fast (max speed is said to be 26 pgs/min) but has a
duty cycle 10 times the LaserWriter (LNZ rated at 40-50K pages/month
the LaserWriter at 4K).
The capacity (duty cycle) of the machine seems to be a major buying
reason. Can anyone think of other reasons -- or perhaps why a handful
of LaserWriters would be preferable?
The Dataproducts LNZ-2660 was scheduled for production deliveries in
November when I sent away for information last September. Now
deliveries are scheduled for January, but the people I talked to said
OEM shipments are going out and there are a few evaluation models
available.
As far as I can tell Dataproduct's LNZ machine (and OEM versions) are
the only PostScript printers near delivery, or being delivered, aside
from the Apple LaserWriter...
- Bill
------------------------------
Subject: Re: Dataproducts LNZ-2660?
Date: 22 Nov 85 08:32:11 EST (Fri)
From: "Christopher A. Kent" <cak@Purdue.EDU>
I haven't seen one here, but I'd love to.
As far as I can tell Dataproduct's LNZ machine (and OEM versions) are
the only PostScript printers near delivery, or being delivered, aside
from the Apple LaserWriter...
We have two QMS 800PS printers here; I don't know under what
arrangement. They look like just about any other Canon-based small
printer, and work about the same. Have the same three-LED controls as
the LaserWriter.
As for the decision about one large printer or many small printers, it
depends on how dispersed your users are. If you can distribute the
printers, it seems worthwhile to have many small printers. If you want
a central output facility, get a big printer; otherwise, it's a pain to
try to balance the load on the gaggle of little guys huddling in a
room. (That's the situation we're in right now.)
I sure hope someone else gets their act together soon. We need a
high-capacity PS printer.
chris
------------------------------
Date: Fri 22 Nov 1985 11:05:24 EST
From: Bob Pellegrino <primerd!bobsun!bob@mit-eddie.MIT.EDU>
Subject: RE:Dataproducts LNZ-2660?
We have a Beta release LNZ-2660 here at Prime. First the bad news...
There is a difference between ENGINE speed and CONTROLLER speed. Yes,
the engine is capable of spitting out 26 pages/minute. But the
controller on this printer cannot go nearly that fast. In fact, in all
but one special case, this printer turns out to be SLOWER than an Apple
laserwriter! The one exception is the case when the text in memory has
already been proccessed and you are simply printing multiple copies of
the same text. (Of course. The controller is already done.) In all
other tests, from simple text to grayscale, the Data Products printer
is slower. In fairness it should be restated that we have a Beta
release version.
However, it is not true that Data Products is the only one making
Postscript printers on the high end. We also happen to have some Beta
printers from QMS (Quality Micro Systems). The people who poisoned us
with that *awful* QUIC language have a full line of Postscript printers
from laserwriter size to a 24 page/minute model.
By the way, is Adobe on the net? I'm surprised not to hear anything
from them...
Bob Pellegrino
Prime Computer, Inc.
ARPA: Pellegrino@bbna
UUCP: decvax!genrad!mit-eddie!primerd!bobsun!bob
------------------------------
Date: Wed 27 Nov 1985 10:41:22 EST
From: Bob Pellegrino <primerd!bobsun!bob@mit-eddie.MIT.EDU>
Subject: Re: RE:Dataproducts LNZ-2660?
> Bob,
> You mentioned you have QMS printers on the "high end" -- these are not
> the PS800 right? Do you know the model number? Is this 24 ppm model
> you mention really 24 ppm or is it another situation where the postscript
> controller cannot keep up and does only as well as a LaserWriter?
> - Bill
> Acting Director of Facilities
> Department of Computer Science
> Columbia University
Bill,
There was never a Postscript version of an Lg2400. The only "high end"
printer that had Postscript was the Lg1200. (12 pages per minute) Since
this printer used the old controller, throughput measurements would be
irrelevant now. It was this printer that they decided to redesign. We
still have an Lg1200a in everyday use though.
Now the plan is to get the new Beta PS800A debugged, and then move it up
to an Lg2400. (not the 1200 first.) For this reason I had us buy
Lg2400's for the time being, with the understanding that they will be
upgraded either for free or for nominal charge next quarter.
By the way, I discovered something interesting last night as I was
printing an 800 (!!) page document on a Laserwriter. Remember that n-up
on a page utility that I posted here last week? Well as it turns out,
the overhead for doing the processing is low enough that the effective
page throughput (at 4-up) is 16 pages/minute. For listings on an 800, that's
pretty good!
I don't know what 16-up gives you, but it is really only appropriate for
document layout proofing at that size anyway.
Bob Pellegrino
Prime Computer, Inc.
ARPA: Pellegrino@bbna
UUCP: decvax!genrad!mit-eddie!primerd!bobsun!bob
------------------------------
Date: Wed 27 Nov 1985 15:26:54 EST
From: Bob Pellegrino <primerd!bobsun!bob@mit-eddie.MIT.EDU>
Subject: RE: Dataproducts printer
A few days ago I posted some speed comparisons between the LNZ-2660 with the
Redstone controller and the Apple Laserwriter. The comparisons were
correct, but what wasn't correct was that the LNZ was a "beta" printer.
In fact it was only an "experimental" printer, not something that they
had intended to ship as a product. Dataproducts has a revised
controller which we hope to see soon.
Bob Pellegrino
Prime Computer, Inc.
ARPA: Pellegrino@bbna
UUCP: decvax!genrad!mit-eddie!primerd!bobsun!bob
------------------------------
Date: 27 Nov 1985 1730-PST (Wednesday)
From: Brian Reid <reid@glacier>
Subject: Printer speeds
Whenever you are looking at the speed of a printer much faster than the
Apple LaserWriter, you must take the speed of the communication link
into account. It is also worthwhile to take the controller architecture
into account. Let me give you an example.
If a printer is being used to emulate a line printer, that means that
it will be printing 63 lines per page and (say) 80 characters per page.
That gives 5040 characters per page. Often pages printed are not full;
let's suppose, in fact, that the average line-length is 40 characters.
That would give us about 2500 characters per average simulated line
printer page.
A 9600-baud link sends roughly 1000 characters per second. It's
actually a bit lower than that, and many small-computer CPUs have
trouble sustaining a 1000-character-per-second output rate over a port
that is not DMA. I know that my Vax 750, even when it's lightly loaded,
has trouble keeping up with 1000 characters/second over a DZ-11 port
when there is other terminal activity going on. Nevertheless, let's use
1000 characters/second as the transfer rate.
It takes 2.5 seconds at 1000 characters/second to ship our sample page
image at this rate.
Now, my measurements of the Apple LaserWriter show me that its
PostScript interpreter can write about 700 characters per second into
the frame buffer if they are all in the cache. At this speed it takes
3.5 seconds per page to fill the frame buffer once all of the
characters are in the machine.
If these two could be overlapped, then 9600 baud would be a very nice
transfer rate for the LaserWriter, because there would be little point
in getting data to the machine much faster than it can be imaged.
However, my understanding of the architecture of the LaserWriter is
that it does not overlap input with page image generation, so in fact
those two times must be added and not maxed. Even when adding them, the
cost of character-writing dominates the cost of communication.
Now let's put a much faster image generator on a printer. In fact,
let's put an infinitely fast image generator on it, so that the speed
at which pages can be printed is limited only by the rate at which data
can be gotten into the printer and the rate at which the frame buffer
can be dumped. If the page could be imaged instantly, and if the frame
buffer could be dumped instantly, the 9600-baud link would limit the
printing speed to 24 pages/minute. However, the page cannot be dumped
instantly, and furthermore, the dumping of the page cannot be
overlapped with the inputting of the next page unless there are two
frame buffers. I don't know the timing info on the Dataproducts
printer, but let's hypothesize that it takes 1/30 of a minute to print
one page, so that a 24 page/minute rate is achieved by printing for
1/30 minute and then 1/120 of a minute to do other things. 1/30 minute
is 2 seconds, and 1/120 minute is half a second. That half a second can
probably be overlapped with the communication time, but the 2 seconds
cannot. This means that unless the controller has 2 frame buffers, a
printer whose hardware can print one page in 1/30 minute can only run
at 60/(2.5 + 2) = 13.33 pages per minute. And that's with an infinitely
fast image generator. If we are willing to grant the image generator 1
second per page of computation time to compute the image (that's mighty
fast), then we are suddenly down to 60/(2.5 + 2 + 1) = 10.9 pages/minute
on this printer whose marking engine and image generator are both 2.5
times faster than the LaserWriter.
If the communication cost could be eliminated completely, the printer
about which we've made these assumptions could be run at 20 pages per
minute (1 second to fill the frame buffer and 2 seconds to dump it; the
1-second fill time can overlap with the half-second "breather" time).
If we kept the 2.5 seconds of communication time but put in two frame
buffers so that they can be double-buffered, then we can print 13.3
pages/minute. If we manage to overlap the communication time and ALSO
use two frame buffers, then we can suddenly jump to 24 pages/minute,
which is the rated speed of the printer. If we manage to reduce the
cmmunication time to under 2 seconds, and also overlap it, and also
have 2 frame buffers, then we can print 30 pages/minute.
All of these numbers that I'm giving you here are just
back-of-the-envelope scribbles; I don't have any real measured data
from real printers. I guess the message I'm trying to get across is
that you can't do much worthwhile discussion of the speed of a
PostScript printer (or any other printer using a frame buffer) unless
you disclose the nature of the communication line with the printer and
the nature of the I/O timing. They also show me that nobody can build a
24 page/minute frame-buffer printer that uses a 9600-baud serial line
unless the I/O can be completely overlapped with computation and also
unless there are 2 frame buffers. Even with those it is going to
require a very fast image generator to keep up.
Brian Reid
Stanford
------------------------------
Date: Sun, 1 Dec 85 14:58:38 est
From: Barry Shein <bzs%buit4%bostonu.csnet@CSNET-RELAY.ARPA>
Subject: Re: Printer speeds
Although I agree with the basic calculations which indicate that
a 9600b serial line will be a bottleneck ultimately in a printer,
Brian Reid's calculations conclude with 'Even with those [ie. more
efficient I/O rates] it is going to require a vary fast image gen-
erator to keep up' (comment mine.) The implication seeming to be
that the technology seems to top out due to various factors at around
24/pages-per-minute, the further implication being such speeds are
not likely to be much of a win as in most applications sustained
throughput will be much lower (do I read too much into this?)
Yet, one cannot deny that, for example, our Xerox8700 (channel attached
to our IBM3081) seems to zip along at around 60 or more pages per
minute, and I hear the 9700 is a bit faster (and, I believe, the
IBM3800 still faster though I suspect a bit dumb about documents.)
The upshot is that unless you are using a very small computer and
a very small link significant increases in speed are already in use
and well within reach, tho admittedly costly (that is, it's not a
matter of technology, it's a matter of price.)
-Barry Shein, Boston University
------------------------------
Date: 1 Dec 1985 1514-PST (Sunday)
From: Brian Reid <reid@glacier>
Subject: Re: Printer speeds
Please read what I said:
"...nobody can build a 24 page/minute frame-buffer printer
that uses a 9600-baud serial line unless the I/O can be
completely overlapped with computation and also unless
there are 2 frame buffers. ..."
This comment has nothing to do with machines like the 8700 that are
attached to a megabit-per-second data channel and that do not have
significant graphics capabilities. I don't mean to denigrate the 8700 and
thee 9700, but they are not graphics printers, they are character printers;
printing graphics is a lot more difficult. (Yes, I know that the 9700 has a
certain limited graphics capability, but is is not at all general).
On the other hand, the 9700 has some fairly impressive performance
statistics: it can print 120 pages per minute, with 8000 characters on each
page. That is a character writing rate of 16000 characters per second, about
22 times the speed of the LaserWriter. However, to achieve that speed, the
9700 used a lot of special-purpose hardware for doing character image
generation; even if the 9700 could print graphics, it couldn't print them at
that kind of speed.
I know that INFO-POSTSCRIPT is for discussing Postscript, not 9700's, but I
think it is important for people to make sure, when they are making
comparisons, that they are comparing apples with apples. A PostScript
printer with the speed of the 9700 would be an impressive beast, indeed. I
shudder to think of the number of frame buffers it would need, the image
generation speed it would need, etc. The 9700 does not eve have a frame
buffer.
Brian Reid
Stanford
------------------------------
Date: Mon, 2 Dec 85 09:17 PST
From: JAlrich.es@Xerox.ARPA
Subject: Re: Printer speeds
I don't think things are quite as bleak as Brian indicates. Brian's
example assumed only one copy per page; as soon as multiple copies per
page are thrown into the equation, the higher print rates become much
more interesting.
//John
------------------------------
Date: Mon 2 Dec 1985 12:04:23 EST
From: Bob Pellegrino <primerd!bobsun!bob@mit-eddie.MIT.EDU>
Subject: Re:Printer speeds
Brian,
You are quite right that raw speed measurements can be misleading if
the speed of the communications link is unknown. Speed
@i(comparisons), however, are still valid given the speed of the
communications link and the buffer sizes are kept constant. The
printer's I/O strategy is fair game, and not to be discounted. (Like
saying: "Car A is faster than Car B, but only because it has a bigger
engine. Right. But Car A is still faster.)
By the way, we are driving our laser printers with 9600 baud serial
lines.
Bob Pellegrino
Prime Computer, Inc.
UUCP: decvax!genrad!mit-eddie!primerd!bobsun!bob
ARPA: Pellegrino@bbna
------------------------------
End of Info-Postscript for Laser Lovers Digest
**********************************************