8 Nov
2017
8 Nov
'17
9:02 p.m.
From: Jon Forrest
In the early days of Unix I was told that it
wasn't practical to write a
pager because such a thing would have to run in raw mode in order to
process single letter commands, such as the space character for going on
to the next page. Since raw mode introduced a significant amount of
overhead on already overtaxed machines, it was considered an anti-social
thing to do.
Something sounds odd here.
One could have written a pager which used 'Return' for each new page, and run
it in cooked mode and not used any less cycles (in fact, more, IIRC the
cooked/raw differences in handling in the TTY driver).
But that's all second-order effects anyway. If one were using a serial line
hooked up to a DZ (and those were common - DH's were _much_ more expensive, so
poor places like my lab at MIT used DZ's), then _every character printed_
caused an interrupt. So the overhead from printing each screen-ful of text was
orders of magnitude greater than the overhead of the user's input to get the
next screen.
IIRC later versions of Unix added the ability to
respond to a specific
list of single characters without going into raw mode. Of course, that
didn't help when full-screen editors like vi and the Rand editor came
out.
Overhead was definitely an issue with EMACS on Multics, where waking up a
process on each character of input was significant. I think Bernie's Multics
EMACS document discusses this. I'm pretty sure they used the Telnet RCTE
option to try and minimize the overhead.
Noel
8 Nov
8 Nov
9:11 p.m.
New subject: pre-more pager?
On 11/8/2017 12:02 PM, Noel Chiappa wrote:
I agree, now that I think about it. I do remember being told not to
write such a thing, due to the load it would cause. Maybe John Nagle,
who had the cube next to mine, might remember better.
From: Jon
Forrest
In the early days of Unix I was told that it
wasn't practical to write a
pager because such a thing would have to run in raw mode in order to
process single letter commands, such as the space character for going on
to the next page. Since raw mode introduced a significant amount of
overhead on already overtaxed machines, it was considered an anti-social
thing to do.
Something sounds odd here. Overhead was definitely an issue with EMACS on
Multics, where waking up a
process on each character of input was significant. I think Bernie's Multics
EMACS document discusses this. I'm pretty sure they used the Telnet RCTE
option to try and minimize the overhead.
I remember ~40 students at UCSB trying to run 'vi' on a Vax 780 with 4MB of
RAM. Between the lack of memory and the overhead of raw mode, it wasn't
a pretty sight.
Jon
10 Nov
10 Nov
6:26 a.m.
New subject: pre-more pager?
On Wed, Nov 8, 2017, at 15:11, Forrest, Jon wrote:
I remember ~40 students at UCSB trying to run
'vi' on a Vax 780 with
4MB of RAM. Between the lack of memory and the overhead of raw mode,
it wasn't a pretty sight.
There's also likely a difference between typing large amounts of text in
raw mode (as vi entails, especially in insert mode) vs issuing single-
character commands.
8:31 a.m.
New subject: pre-more pager?
On Fri, Nov 10, 2017 at 12:26:30AM -0500, Random832 wrote:
On Wed, Nov 8, 2017, at 15:11, Forrest, Jon wrote:
I also had to do programming assugnment on an overloaded VAX.
Luckily you can switch to ex mode in vi. ex is much easier on the
macine since it doesn't work in raw mode.
-Otto
I remember ~40 students at UCSB trying to run
'vi' on a Vax 780 with
4MB of RAM. Between the lack of memory and the overhead of raw mode,
it wasn't a pretty sight.
There's also likely a difference between typing large amounts of text in
raw mode (as vi entails, especially in insert mode) vs issuing single-
character commands. 
8 Nov
8 Nov
9:18 p.m.
New subject: pre-more pager?
On Wed, Nov 8, 2017 at 8:02 PM, Noel Chiappa <jnc(a)mercury.lcs.mit.edu>
wrote:
If one were using a serial line
hooked up to a DZ (and those were common - DH's were _much_ more
expensive, so
poor places like my lab at MIT used DZ's), then _every character printed_
caused an interrupt.
Right, although many (most) Unix sites used Able DHDM's which were cheaper
than DEC DZ's, took less space in the backplane, had full modem control
(inc RTS/CTS handshaking in HW) and were DMA. But, my memory is that they
did not show up until '78 though.
The truth is DZ sucked from a performance standpoint. Besides not being
able to support RTS/CTS flow control, the 8 serial ports shared an input
buffer in the HW IIRC and the UART only had 2 chars of buffering. An
8-port DZ with all lines running 19.2K baud could kill a 780. People
that ran the "Berk-net" (which a pre-TCP networking system for UNIX that
ran on serial lines) needed spread out the Berk-Net load between different
DZ lines. No so if you have Able gear.
So I'll take the 'common' note to be if you used a 'pure DEC' set
up - you
used DZ-11s. But as you said most of us were looking at costs, so
'foreign' peripherals were also 'common' in the UNIX community.
DH11's
were preferred.
9:28 p.m.
New subject: pre-more pager?
We actually still had some real DEC DH’s on our system. These are what the DHDM
emulates, but are implemented in its own backplane full of various flipchip cards.
At least the DZ doesn’t loop on the ready bit like the kernel printf (not to be used for
idle chit-chat)..
From: TUHS [mailto:tuhs-bounces@minnie.tuhs.org] On Behalf Of Clem Cole
Sent: Wednesday, November 8, 2017 3:19 PM
To: Noel Chiappa
Cc: TUHS main list
Subject: Re: [TUHS] pre-more pager?
On Wed, Nov 8, 2017 at 8:02 PM, Noel Chiappa <jnc(a)mercury.lcs.mit.edu> wrote:
If one were using a serial line
hooked up to a DZ (and those were common - DH's were _much_ more expensive, so
poor places like my lab at MIT used DZ's), then _every character printed_
caused an interrupt.
Right, although many (most) Unix sites used Able DHDM's which were cheaper than DEC
DZ's, took less space in the backplane, had full modem control (inc RTS/CTS
handshaking in HW) and were DMA. But, my memory is that they did not show up until
'78 though.
The truth is DZ sucked from a performance standpoint. Besides not being able to support
RTS/CTS flow control, the 8 serial ports shared an input buffer in the HW IIRC and the
UART only had 2 chars of buffering. An 8-port DZ with all lines running 19.2K baud
could kill a 780. People that ran the "Berk-net" (which a pre-TCP networking
system for UNIX that ran on serial lines) needed spread out the Berk-Net load between
different DZ lines. No so if you have Able gear.
So I'll take the 'common' note to be if you used a 'pure DEC' set
up - you used DZ-11s. But as you said most of us were looking at costs, so
'foreign' peripherals were also 'common' in the UNIX community.
DH11's were preferred.
11:47 p.m.
New subject: pre-more pager?
On Wed, 8 Nov 2017, Noel Chiappa wrote:
But that's all second-order effects anyway. If
one were using a serial
line hooked up to a DZ (and those were common - DH's were _much_ more
expensive, so poor places like my lab at MIT used DZ's), then _every
character printed_ caused an interrupt. So the overhead from printing
each screen-ful of text was orders of magnitude greater than the
overhead of the user's input to get the next screen.
Looxury! We had to make do with DJ-11s...
--
Dave Horsfall DTM (VK2KFU) "Those who don't understand security will
suffer."
2567
days inactive
2569
days old
6 comments
7 participants
participants (7)
-
Clem Cole -
Dave Horsfall -
Forrest, Jon -
jnc＠mercury.lcs.mit.edu -
Otto Moerbeek -
Random832 -
Ron Natalie