[Simh] PIP10 on PDP-8 SIM

[Johnny Billquist]

On 2013-03-19 14:56, Timothe Litt wrote:
>
> This communication may not represent my employer's views,
> if any, on the matters discussed.
>
> Consider a timesharing system, with 4 users each trying to transfer from
> their own DTA.  With a TC11, you serviced user 1, user, 3, user 4, user
> 2 sequentially.   The controller was tied up for the whole seek for each
> user - potentially from one end of the tape to the other.

Oh. You are talking about overlapped seeks. I thought you meant 
something else.

I just checked the TC01 controller (for the PDP-8) and it certainly can 
do this too. But you need to programatically time things in this case, 
to go back to the drive when you think it is getting close, and start 
fiddling. Also checked the TC11, and it too can run several tape units, 
as long as you reselect the tape drive when you think you actually want 
to start doing any operations. Both can have tapes continue to run in 
either direction while doing operations on another tape.
Sounds like exactly the same deal as with the TD10.

The TD8E, which I mentioned being the most stupid controller on earth, 
requires this solution all the time, as there is no actual seek, nor is 
there any DMA. You either poll the controller all the time, or else you 
setup a time interrupt when you think it is appropriate to start polling.
(Of course, OS/8 just polls the darn tape the whole time...)

> The TD10-'s dead reckoning allowed  the seeks to proceed in parallel.
> E.g. we'd start all 4 seeks, allowing the drives to run unsupervised by
> the controller.  When the shortest seek was near completion, the driver
> would tell the controller to pay attention to that drive ('connect'), do
> the transfer, and stop.  Then the next.  Besides being impressive to
> watch (8 drives spinning in various directions), the user request
> latency was reduced.   (Some systems had two controllers, for 16 drives!)

I don't know if any driver for any PDP8 OS, or PDP11 OS ever did 
overlapped seeks, but the hardware can definitely do it.
The one OS I can tell about is RSX, and it did not support overlapped 
seeks on TC11. (And of course I know OS/8, but the answer there is that 
nothing ever did overlapped seeks on anything, since all I/O was polled 
anyway.)

> The KA, KI and KL (with the IO Bus adapter) all used the TD10
> controller.  The -6 had another controller - type 551.  I think it was a
> dumb controller, but didn't use it personally.

Ok.

> 129 x 12 = 1548 bits/block; 256 x 18 = 4608 bits/block ; 2 formats. Gory
> details of the TD10 are at

Oh. You just meant the number of bits in a block. Well, you can do any 
size you want, just as you can do any number of blocks you want, withing 
physical limits, and probably some other restrictions that I can't think 
of right now.

> Since PIP10 (the pdp-8 reader for pdp-10 tapes) was the question, the
> format info I provided was for the 10.  TOPS-10.  MAC had another format
> for the -6.  TOPS-20 didn't officially support DECtapes, but internally
> we had systems with them.  They used the PDP-10 format.

Makes sense.
Anyway, yes, PIP10 would parse the format you described, once it 
actually manages to read the blocks. That is the tricky detail that 
seems to not be working...

> Much of the pdp-10 format documentation is a mixture of octal and
> decimal.  I picked decimal because these days, thinking in octal is a
> (mostly) lost art.

:-)

> IBM and many other vendors used 36-bits, because the (BCD) character
> codes of the day were 6-bits, so it packed nicely.  The -8 was 12 bits
> for the same reason, only it was cheaper to build.

I don't know how much the PDP-8 (or rather, PDP-5) was 12 bits because 
of any BCD issues. But it was cheap.

> The PDP-10 supported arbitrary byte sizes (1-36 bits), so 7-bit ascii,
> 8-bit ascii or 9-bit ITS emacs encodings were all easy to deal with.

I know. The PDP-10 is a nice architecture.

> The -11 was designed after 8-bit character codes became popular. EBCDIC
> was (and is) an ugly mess; ASCII cemented 8-bit character codes.  And
> machines were then built as multiples of 8 bits, 8, 16, 32, 64 being the
> most popular.

I can only agree about EBCDIC. ASCII wasn't really 8-bit though. But 
7-bit is close enough.

	Johnny

>
> On 19-Mar-13 09:25, Johnny Billquist wrote:
>> On 2013-03-19 14:03, Timothe Litt wrote:
>>>> The DECtape format as such, with all the headers and so on, is the
>>>> same on all tapes. A normal PDP-8 formatted tape will have 129
>>>> (12-bit) words, however, while a PDP-10 (or any other 18-bitter) would
>>>> have 128 18-bit words (if I remember right).
>>> Pretty much right.  129 may be slightly misleading.  The format is 129,
>>> but the -8 used  only 128 of the 129 12-bit words/block for data.
>>
>> Right. :-)
>>
>>> The PDP-10 is 36-bit, not 18-bit.  A PDP-10 (actually, non-PDP-8)
>>> formatted DECtape would have 578 blocks of 128 36-bit words. (256
>>> 18-bit words at the hardware level.)
>>
>> I know that the PDP-10 is 36 bits. :-) I was talking about the DECtape
>> as such, which physically always stores 18 bits. A PDP-10 would
>> obviously store its 36-bit words using two DECtape words. Any 18-bit
>> machine is straight forward, while the PDP-11 (as you mention below)
>> only use 16 bits of the 18-bit word.
>>
>> I thought it was 128 18-bit words, but I may well remember that wrong,
>> and the "normal" format is 256 18-bit words. Makes sense, when I think
>> about it.
>>
>> The number of blocks is not absolutely fixed, but 578 is the
>> "standard" except on the PDP-8, I guess.
>>
>>> Blocks 0, 1, 2 are for DTBoot (hardware read-in bootstrap), and didn't
>>> contain user data.
>>
>> That would be very depending on which machine we're talking about, and
>> which controller... :-) But I assume you're talking about the PDP-10
>> here. Did all PDP-10 with DECtapes use the same controller? I assume
>> both the KA and KI had DECtapes, as well as the PDP-6.
>>
>>> Block 100 is the directory block.  Thus 574 blocks for user data. The
>>> directory holds up to 22 files, plus a map of which file owns each
>>> block.
>>>
>>> The user data blocks have a one word header (LH = next block of file; RH
>>> = first block & words used in this block) + 127 words of payload.  This
>>> differed from disk files, where all 128 words were payload, so
>>> inattentive programmers could make a number of mistakes.  (E.g. random
>>> access block isn't word/128; you had to pay attention to the buffer's
>>> word count, etc.)
>>>
>>> Data blocks of a file are (usually) not contiguous; this allowed the
>>> drive to stop and restart while reading or writing without having to
>>> reverse direction.  The spacing depended on what blocks were free when
>>> files were written, and the data mode.  (Files written in 'core dump'
>>> modes were assumed to be read by the monitor without stopping, so the
>>> gap was smaller, allowing larger programs to be read without reversing
>>> direction.)
>>>
>>> The gory details of the format are in the Monitor Calls manual
>>> (TOPS-10).
>>
>> TOPS-10 DECtape format is definitely not something I know anything
>> about. But thanks for the rundown.
>>
>>> The PDP-11 used 18-bit format, but ignored the high 2 bits of each word
>>> (except when reading PDP-10 tapes; the hardware for that was tricky as
>>> the high bits had to be read with programmed IO; the low 16 were DMAed
>>> on the TC-11.)
>>
>> Yes! Thanks for bringing another murky memory.
>>
>>> The low-level formatting that established the mark track, end zones and
>>> block delimiters was done via a stand-alone diagnostic. This differs
>>> between the two formats.  The directory block could be initialized under
>>> timesharing.
>>
>> Um? What two formats?
>>
>>> Directory structure given is for the PDP-10; other OSs used different
>>> formats.
>>>
>>>  From an emulation point of view, the PDP-10's controller was the most
>>> interesting; the driver does dead reckoning; that is, it will start a
>>> drive spinning for a seek, disconnect, service other drives, and
>>> reconnect just before the desired block is expected to be over the read
>>> head.  So real-world timing matters.  The other controllers (and thus
>>> OSs) didn't support this.
>>
>> Wow. The TC08 actually did seeks while actually looking at the tape,
>> and then did DMA as well. So you didn't have to pay any more attention
>> than to a disk controller.
>> Essentially, you asked it to seek, and it indicated when the seek was
>> done. You asked it to read, and it read.
>> As far as I can remember, the TC11 is the same.
>> Not sure what you mean by "other controllers didn't support this".
>> What is "this". Doing dead reckoning? Why would you actually want to
>> do dead reckoning. It makes much more sense to actually look at the
>> tape the whole time it is spinning by, and not only when you are
>> getting close to where you want to be. And the hardware can do that
>> without your involvement.
>>
>> In a way, however, I'd say that the TD8E is the most "interesting"
>> controller to emulate, since it is so incredibly stupid that you
>> actually have to emulate the low level format of the tape for that
>> controller to work.
>> It is truly ugly. And that controller was a headache if you ever
>> wanted to use it in a timesharing operating system, since it did not
>> do DMA, it did not really do seeks, or in fact anything at all except
>> just feed the bits from the tape to the computer as they whizzed by.
>>
>>> Oh, all numbers above are radix 10.
>>
>> What's wrong with you. ;-)
>>
>>> The link for OS8 that I posted yesterday was via filewatcher; the direct
>>> link
>>> isftp://sunsite.unc.edu/pub/academic/computer-science/history/pdp-8/os8/
>>> Sorry if this was confusing.
>>>
>>> Of potential interest to low-level folks is
>>> http://patft.uspto.gov/netacgi/nph-Parser?Sect2=PTO1&Sect2=HITOFF&p=1&u=/netahtml/PTO/search-bool.html&r=1&f=G&l=50&d=PALL&RefSrch=yes&Query=PN/3387293
>>>
>>>
>>> Hope this is useful.
>>>
>>> This communication may not represent my employer's views,
>>> if any, on the matters discussed.
>>
>> Thanks for all the interesting bits and pieces.
>>
>>     Johnny
>>
>>>
>>> On 18-Mar-13 16:26, Johnny Billquist wrote:
>>>> On 2013-03-18 17:44, Bob Supnik wrote:
>>>>> I was trying to get a debug setup for PIP10, per Ian King's mail,
>>>>> when I
>>>>> discovered that none of my OS/8 images have PIP10 on them. This
>>>>> certainly explains why the feature has never been tested before. I
>>>>> suspect that ReadAll and WriteAll either are not working at all, or
>>>>> are
>>>>> not working when the DECtape format is 18b. Another possibility is
>>>>> that
>>>>> PDP-10 DECtape format is not the same as 18b format, at the
>>>>> nitty-gritty
>>>>> level (format of headers and trailers).
>>>>
>>>> The DECtape format as such, with all the headers and so on, is the
>>>> same on all tapes. A normal PDP-8 formatted tape will have 129
>>>> (12-bit) words, however, while a PDP-10 (or any other 18-bitter) would
>>>> have 128 18-bit words (if I remember right).
>>>> The PDP-8, when doing 12-bit formatted tapes, just packs data in a way
>>>> that is rather different from an 18-bit machine. But at the tape as
>>>> such, there is nothing odd about it.
>>>>
>>>> But I can see lots of potential for errors when emulating this whole
>>>> thing.
>>>>
>>>> I couldn't give exact details on lots of bits without looking in
>>>> manuals, but in essence a DECtape is always doing 18-bit words. That
>>>> is done by doing 6 groups of 3 bits each.
>>>> A PDP-8 will pack three 12-bit words into two 18-bit words. This means
>>>> that a DECtape block for a PDP-8 will only have 86 18-bit words.
>>>> So the blocks are shorter, but you have more of them, when the tape is
>>>> formatted for a PDP-8.
>>>>
>>>> I hope (assume) that you already know all of this. If not, let me
>>>> know, and I can try helping out some more.
>>>> I actually did dump a few 18-bit tapes on my PDP-8 only a few months
>>>> ago, which is when I actually had to dig rather deep into all of this.
>>>> PIP10 was one of the things I really looked into. But since my tapes
>>>> had actually been written on a PDP-15, I had to write my own code in
>>>> the end, to just dump the raw data.
>>>>
>>>>> If anyone has a canned OS/8 V3C image with PIP10, please post it
>>>>> somewhere (like Mediafire) and let me know by email.
>>>>
>>>> I think someone already posted this, but I know I have that software,
>>>> including sources (if I remember right) somewhere. Let me know if it
>>>> can't be located anywhere else.
>>>>
>>>>     Johnny
>>>>
>>>> _______________________________________________
>>>> Simh mailing list
>>>> Simh at trailing-edge.com
>>>> http://mailman.trailing-edge.com/mailman/listinfo/simh
>>>
>>>
>>>
>>> _______________________________________________
>>> Simh mailing list
>>> Simh at trailing-edge.com
>>> http://mailman.trailing-edge.com/mailman/listinfo/simh
>>>
>>
>>
>
>
>
>
> _______________________________________________
> Simh mailing list
> Simh at trailing-edge.com
> http://mailman.trailing-edge.com/mailman/listinfo/simh
>