[continued from previous message]

The interaction between the delay lines and vertical refresh is
somewhat complicated. The vertical refresh time was designed to be
exactly the same as the 5.5545 ms time it took a buffer to fully
circulate, while the time to display a vertical scan line was exactly
twice this time. Two buffers were interleaved to provide the vertical
scan lines. During the first time interval, the first buffer provided
pixels for the top half of the line. During the second time interval,
the second buffer provided pixels for the bottom half of the line.
The third time interval was used for vertical refresh. This pattern
continued until the end of the buffers, so every third slot in a
buffer was displayed while the "unused" pixels were recirculated.
This process was repeated three times, offsetting the start point in
the buffer, so the buffers were displayed entirely.

[18]
Another curious feature of the IBM 2260 display is how it converted
the 6-bit character code into the 5x7 block of pixels representing
the character. It used a special core memory plane that only had
cores for 1 bits and omitted cores for 0 bits, so it acted as a
read-only memory. The result is you could actually see the characters
in core plane, as illustrated below. The core plane holds nine 7-bit
words for each of the 64 characters: the first five words held the
pixel block, while the four other words were a lookup table to
convert the EBCDIC character code (2848 code) to or from ASCII or a
tilt-shift code used to control the Selectric-like printer (Model
1053).

<http://static.righto.com/images/terminal/core-plane-w400.jpg>

Part of the character generation core plane, showing the segment for
the character 'A'. The diagonal lines indicate ferrite cores; I've
colored the cores storing the character image. The core plane was a
72x56 grid in total representing 64 characters. Image based on 2260
Field Engineering Theory of Operation Manual p2-82.

<http://www.bitsavers.org/pdf/ibm/2260/
Y27-2046-3_2260_2848_FETOM_Mar68.pdf>

[19]
IBM apparently liked the idea of splitting display hardware between
two users, because they did that with the IBM 3742 Dual Data Station
(1973). This system let two operators enter data onto 8" floppy
disks. The bizarre part is that it had a single vertically-mounted
CRT display. The small black box in the middle of the desk is a pair
of mirrors that showed half the screen to each operator. The result
was a very squat display with just three lines of 40 characters,
enough for a status line and 80 characters of data.

<https://en.wikipedia.org/wiki/IBM_3740>

The IBM 3742 Dual Data Station allowed two operators to type data
onto floppy disks. Image from IBM 3740 System Summary.

<http://static.righto.com/images/terminal/ibm-3742-w400.jpg>

<http://bitsavers.org/pdf/ibm/3740/GA21-9152-0_IBM_3740_ DataEntrySystem_SystemSummary_and_InstallationManual_ PhysicalPlanning_Jan73.pdf>

[20]
The lines of text in the screenshot appear closer together than
double-spaced, even though they are double-spaced. The reason is that
the dots on the screen are a bit larger than one pixel, so they
encroach into the space between the lines. In other words, the
display alternates 7 lines of character pixels and 7 blank lines, but
it looks more like 9 lines of character pixels and 5 blank lines.

[21]
Televisions and CRT displays normally use a raster scan, scanning the
electron beam across the screen in horizontal scan lines, making a
series of lines from top-to-bottom. The 2260, on the other hand, has highly-unusual vertical scan lines; the scan lines are top-to-bottom,
and the series of lines progressed left-to-right across the screen. I
haven't been able to determine any reason why the 2260 has vertical
scanlines. I assume it made the timing work out better somehow.

<https://en.wikipedia.org/wiki/Raster_scan>

[22]
Here are my calculations on the maximum number of lines that could be
displayed by the 2260. A 250 nanosecond pixel rate and 30 Hertz
refresh give a maximum of 133,333 pixels that can be displayed on the
screen. If each character is 6x7 pixels and there are 80 characters
per line, 39.7 lines could be on the screen. Vertical refresh takes
1/3 of the time because of interaction with the delay lines, [17]
dropping this to 26.5 lines. Because the 2260 splits pixels across
two displays, that yields at most 13.25 lines on the display,
ignoring horizontal refresh. Therefore, 12 lines of text are about
what the hardware could support. (I should point out that it's
possible they decided on 12 lines first and selected the other design characteristics to fit this.) Note that the next reasonable line size
would be 16 lines. The low-end model displayed 6 lines of 40
characters (i.e. 3 punch cards), so the next step for it would be 8
lines of 40 characters (four punch cards). Since the high-end model
uses four buffers, that would yield 16 lines. The point is that it
would have been a large jump to go beyond 12 lines.

[23]
The 2260 came in three models. Model 1 displayed 6 lines of 40
characters. Model 2 displayed 12 lines of 40 characters. Model 3
displayed 12 lines of 80 characters. The main difference in
implementation was that they used 1, 2, and 4 buffers respectively.
The 40-character models refreshed at 60 Hz rather than 30 Hz, since
they had half the (vertical) scanlines.

[24]
The aspect ratio of the IBM 2260's text was very different from the
screen's aspect ratio. With the bezel, the screen's useful display
area is 9.5 by 5.7 inches (5:3 ratio). Note that the aspect ratio of
the text was very different from a standard 4:3 ratio. The 40x6
display format is 6.5 by 2.25 inches (almost 3:1 ratio). The 40x12
display format is 6.5 by 4.5 inches (a bit over 4:3 ratio). The 80x12
display format is 9 by 3 inches (3:1 ratio). Information on the
2260's screen size is in the FE Manual chapter 2.

<http://www.bitsavers.org/pdf/ibm/2260/
Y27-2046-3_2260_2848_FETOM_Mar68.pdf>

[25]
The 1974 Datapro report gives the price for the IBM 2260 system as
$1270 to $2140 for the display unit and $15,715 to $86,365 for the
controller. The IBM 3270 in comparison was $4,000 to $7,435 for the
display unit (3277) and $6,500 to $15,725 for the controller. Note
that compared to the 2260, the 3270 moved much of the complexity from
the controller to the display unit, which is reflected in the prices.

<http://www.bitsavers.org/pdf/datapro/datapro_70/ 70D-010-20_All_About_CRT_Display_Terminals_Apr1974.pdf>

[26]
The IBM 3270 was a line of terminals. Like the 2260, itonsisted of a
3271 or 3272 Control Unit (interesting video here) along with the
terminals (3275 or 3277 Display Stations). These could display 40x12
or 80x24. For simplicity, I'll refer to the whole system as the 3270.
Over the years, IBM introduced more models in the 3270 line,
including color and graphics terminals, supporting lower case as well
as display sizes such as 80x32, 80x43, and 132x27. The 3270 PC (1983)
was an enhanced IBM PC that acted as a 3270 terminal. However, I'm
going to focus on the original 3270 terminals, since those had the
most influence.

<https://www.youtube.com/watch?v=7VHHj8P01Xw>

<https://en.wikipedia.org/wiki/IBM_3270_PC>

[27]
A 480-bit shift register might seem like a strange size, since it's
not a power of two. However, since shift registers don't have address
bits, they can be arbitrary sizes. For instance, Collins made dual
66-bit shift registers, to support 64-bit data plus 2 parity bits.
Fairchild made 480-bit shift registers for CRT displays. 500-bit
shift registers were built "to operate in equipment where storage
lengths in 100 bit multiples are required." Texas Instruments built
dynamic bipolar shift registers in 253-bit, 349-bit, and 501-bit
sizes which were useful for Digital Differential Analyzers. The point
is that shift registers can be built in arbitrary sizes, so there is
no need to use a power of two.

<https://archive.org/details/ bitsavers_alliedElecAlliedElectronicsCatalog_133567683/page/n69>

<https://www.tinaja.com/glib/shiftreg.pdf>

<http://smithsonianchips.si.edu/ice/ICE_LSI_DRA1969_1970s11.pdf>

Schematic symbol for the 480-bit shift register in the 3270. Inputs
are data and the two-phase clock. "SPEC" indicates a special circuit.
From the ALD, page MP151.

<http://static.righto.com/images/terminal/spec-w200.jpg>

<http://bitsavers.org/pdf/ibm/3270/fe/3272_ALD_LC_YF_Vol_1_Apr75.pdf>

The 3270 used banks of ten 480-bit shift registers to store 480
10-bit data words (9 bits and parity), unlike the earlier 2260 delay
lines, which stored pixels.

[28]
Software support for the 3270 included DIDOCS (Device Independent
Display Operator Console Support), using the 3270 as a mainframe
system console; VIDEO/370 (Visual Data Entry Online), a program that
allowed customers to design forms for data entry; DATA/360, a program
that emulated an IBM 29 card punch, but provided editing and
validation; IMS (Information Management System); (CICS) Customer
Information Control System, which allowed interaction with a
database; IQF (Interactive Query Facility), another database system;
and TSO (Time Sharing Option).

[29]
The 132-column width for terminals was motivated by the ubiquity of
IBM's 132-column printers.

<http://www.righto.com/2019/01/accounting-machines-ibm-1403-and-why.html>

The DataMaster's influence on the IBM PC is described in two articles
by Dr. Dave Bradley: The creation of the IBM PC in Byte, Sept. 1990;
and A personal history of the IBM PC, IEEE Computer, Aug 2011
(paywalled). The Wikipedia article DataMaster System/23 also provides information.

<https://archive.org/stream/byte-magazine-1990-09/ 1990_09_BYTE_15-09_15th_Anniversary_Summit#page/n451/mode/2up>

(paywalled article)

<https://www.computer.org/csdl/magazine/co/2011/08/mco2011080019/
13rRUwInv7Y>

<https://en.wikipedia.org/wiki/IBM_System/23>

[31]
Dr. Bradley explained that the designers of the IBM PC weren't
concerned with compatibility with other systems. For instance, you
might expect the IBM PC to be compatible with the 3270 terminal.
However, the IBM PC's keyboard had 10 function keys while the IBM
3270 terminal had 12. This incompatibility was finally fixed with the
IBM PS/2 keyboard (1987).

[32]
To confirm the popularity of 80x24 terminals versus 80x25 terminals,
I took a look at the GNU termcap file. I counted and found there were
over 5 times as many 24-line terminals as 25-line terminals, and the
25-line terminals were mostly PC-based. 80-column terminals were over
5 times as popular as 132-column terminals, the runner-up.

<https://ftp.gnu.org/gnu/termcap/>

From:
<http://www.righto.com/2019/11/ibm-sonic-delay-lines-and-history-of.html>

Comments
========
Michael Leis said...

Funny story: While we were demonstrating a breadboard of the VT00 to
Gordon Bell, CTO, he asked me if we could make the screen display a
132 character line to match the output of line printers.

I said I would look into it, but thought to myself that with the
resolution we had on-screen it would be unreadable. Thank goodness, I
did not say that to Gordon.

I did some character bit sketches that reinforced my feeling, but to
be sure I went to the lab to see if I could simulate it. Since I had
designed the tv display, I knew how to shrink the horizontal
deflection to allow 80 characters to display in the space they would
be on a 132 character line. And I found they were completely legible!
Spec changed. Our brains do a good job of recognizing fuzzy characters.

In hindsight, this turned into a big profit for Digital. Since we
were under cost pressure we did not put in enough memory to display a
24x132 screen in the base VT100, but we sold an add on a memory
card at somewhat high-price. Most people bought the add-on.

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)