DHV (was: RL01 drive select plug and power supply questions)

Allison ajp166 at bellatlantic.net
Sat Oct 22 13:56:59 CDT 2005


>
>Subject: Re: DHV (was: RL01 drive select plug and power supply questions)
>   From: Johnny Billquist <bqt at Update.UU.SE>
>   Date: Sat, 22 Oct 2005 18:47:02 +0200 (CEST)
>     To: cctalk at classiccmp.org
>
>On Fri, 21 Oct 2005, Allison <ajp166 at bellatlantic.net> wrote:
>
>> >Subject: Re: RL01 drive select plug and power supply questions
>> >   From: Paul Koning <pkoning at equallogic.com>
>> >   Date: Fri, 21 Oct 2005 10:58:55 -0400
>> >     To: cctalk at classiccmp.org
>> >
>> > Allison> Yes I do.  However for a single user system the load is not
>> > Allison> an issue.  If your running a timeshare system such as RSTS
>> > Allison> or RSX with more than one user then DHV11 sense as well.
>> >
>> >Not true.  9600 baud is 960 interrupts per second, on a character I/O
>> >device.  That's a big number for a PDP-11.  Output will be
>> >significantly less of a burden with a DH type output controller than
>> >with other types -- even for just one active terminal.
>>
>> Depends somewhat on the OS.  It dont know about you but most people
>> can't type much faster than 100WPM (less than 10 chars sec).
>
>Um, that's irrelevant. A DH type controller also interrupt on each
>character input. It's output that differs. And filling a full screen is
>1920 characters, which at 9600 bps will take about 2 seconds. Do that on
>one terminal will the system will definitely notice. If you have a serial
>printer of some speed (such as an LN03) we're talking about a lot more
>than about 2k of data sometimes. So even with a single user system, it can
>make a big impact. And of course, if you have anything else running at the
>same time, it will make it even worse. Multi-user systems definitely
>suffer if you use DL11 controllers for users. It's more or less a no-no.

Results differ.  Actualy outputting a full screen to a vt100 at 9600
causes a fair number of input chars (Xon/Xoff buffer management) 
and printers (la100, LN01, LN03 all have buffers larger than 2k
before they assert buffer controls (be they hardware or XON/XOFF).
There is a distinct differnce in system performace due to those 
factors.

>I'm not even sure you can drive one single line at 9600 bps at full speed
>on a DL11, let alone 19200.

LSI-11/PDP-11/03 that is tue due to processor speed.  By the 11/23B
38.4 works!  (I run a TU58 that way)

>> For RT11 it's mostly unimportant.  For unix (the most sensitive
>> to interrupt loading you _may_ care as a single user).
>
>It matters for both, as I've pointed out above.

In theory, in the practical world there are systems where idle cycles
are a reality.  You have to use all of those up first.

>> > Allison> For most of my 11s four lines is the limit for what I can
>> > Allison> seem to keep busy.  Figure a user terminal, LA100 Printer
>> > Allison> and serial line for modem or data line to another system. At
>> > Allison> the extreme I've run two terminals for OSs that support that
>> > Allison> but, I can only type on on at any instant. ;)
>> >
>> >Sure, if you're mostly doing editing, then the CPU burden of high
>> >speed output may not be obvious.  If you had an LN03 or similar
>> >printer, you might see it more easily.  An LA100, of course, isn't
>> >much of a problem because it is quite slow.
>>
>> Actually printers are a bursty load (fill the buffer and go away)
>> and I've found that in practice the faster you fill the buffer
>> the better (high line rates or use a parallel interface).
>
>Yes, and that burst will drop a PDP-11 to it's knees if it's on a DL11 at
>high speed. Sure, if you're running 2400 bps then you'll live. But not
>many does these days.

Note the bursty loads were felt worse (incresed delays) when the baud 
rate was medium (below 4800).  Too much "barely thinking" inbetween
interrupts waiting to fill the buffer on the printer.  With the LPV11
(same problems as DL only very fast parallel byte transfer (
20-40,000bytes/S) the driver would push data from its buffer till 
the printer yelled wait. So the time impact was actually lower.  
The same driver would push a DL the same way for a LA100 or 
LN03 as both had big buffers and would take it fast as you could 
send it (9600!) and then send back XOFF to stem the tide and the 
system would return from the printer driver.  Larger print jobs 
had a differnt impact but as printer speed increased their buffer 
sizes helped to absorb that.  Keep in mid this was an office system
so print jobs were typically 1-4 pages (small).

Printing has a different interrupt load then interactive terminals.
Interactive terminals keystrokes are slow and screen fills are the 
real load.

Define PDP-11.  The family of CPUs and performance varies greatly.
For example the 11/44 could not take a TU58 at 9600, (barely at 4800).
Yet I run a BA11-VA with 11/23, M8059 ram, DLV11J and MRV11 with
a tu58 as boot and mass storage under RT11XM.  works fine at 38.4K 
baud. The only slowness there is waiting for the TU58 to read the 
directory then seek to the file. 

>> Performance was good enough that we used that over the VAX
>> (line lengths limited us to 2400baud in our part of the mill).
>
>Yes, interrupt performance on the VAX was not pretty. One or two DZ11 
>on a VAX-11/750 would kill it.

Yes but the limiting factor there was not the VAX interfaces but 
the greater than 3000ft of wire between us and the VAX. ML3-6 to
ML1-1 was a long walk!  Ex mill rat here!

However 750 performance was pretty tepid, the DMF32 was only ok.
When we moved up to it a few complained it was slower than the 
11/23B but easier to use.  Later on the whole mess was solved
with multisessioning on LAT and LAT servers. 


Allison



More information about the cctalk mailing list