Talk:ITS machine configurations

AI PDP-6

I'm pretty sure that by the time I got to Tech Sq (in 1977), the PDP-6 was no longer in use. It wasn't physically removed until some years later, though. Jnc (talk) 15:43, 14 December 2017 (CET)

Right. I have collected all information I have found on each of the classical ITS machines here: [1], but I figured it was too much for this page. User:Larsbrinkhoff 18:14, 14 December 2017 (CET)

Should we have an ITS machines configurations page? (The name is just a quick first hack; suggestions for something better sincerely requested!) Looking at the recent 9th floor pictures from the CHM, we now have even better data (especially on MC)! It's too bad that collection doesn't include any of AI, though. Jnc (talk) 18:45, 4 February 2024 (CET)

Yes, that could be good to have. The CONFIG file is a good source too. I'm particular to the AI KA10 myself, which has lots of interesting hardware.

AI is so photo shy! I have found photos and video from various places, but barely anything capturing AI. Just the odd corner here and there. Larsbrinkhoff (talk) 19:27, 4 February 2024 (CET)

MC configuration

So I'm trying to work out MC's configuration. (I know, I know, I could look in CONFIG >, but I'm lazy.) Here's what I remember/can work out:

• 8 MF10's (originally); later upgraded to MH10's.
• (later) An Ampex ARM-10L 'external' memory box was added
• 3 RP04's
• So, therefore, some sort of MASSBUS controller, but since it was a KL10 Model A, not an RH20; probably an RH10
• That would have needed a DF10
• A TM10 of some sort; dunno if that had a separate DF10, or shared the one the RH10 used
• Some sort of high-end DEC magtape drive
• A DL10 with a PDP-11/40
• Later, first one, and then a second Trident drive, attached to the DL10 PDP-11
• The front-end -11 had a bunch of serial lines, probably a DH11
• CHAOSNET eventually on the DL10 PDP-11
• I'm not sure why they initially acquired the DL10+PDP-11 (i.e. what it did to begin with); probably more serial lines

Well, that's a start. Jnc (talk) 17:09, 12 April 2021 (CEST)

According to SYSDOC;POOR MC, the tape drive was a TU40 or TU41 (probably TU40). Also, it seems from things said there that it did have its own DF10. Jnc (talk) 17:20, 13 April 2021 (CEST)

According to this, the TM10 had a DF10 (a TM10 by itself did not connect to a PDP-10 Memory Bus. Jnc (talk) 01:23, 26 October 2022 (CEST)

That is consistent with CONFIG >. MC and DM had TM10B with DF10. AI and ML had TM10A using IO bus only. (ML had a DF10 apparently only for the RP10.) Larsbrinkhoff (talk) 06:57, 26 October 2022 (CEST)

POOR MC might be confused. According to this it was a TU41. Which makes sense, because according to the decsystem10 Configurator, the TU40 was the 9-track version, and the TU41 was the 7-track, and I'm pretty sure it was a 7-track drive. Maybe it had a TU40 label on it, because DEC was too confused? Jnc (talk) 22:40, 6 February 2024 (CET)

I have tried to summarize changes to CONFIG > for all machines over the years: https://github.com/PDP-10/its/issues/1168

Most of what you wrote seems spot on. I can see there was a third T-300. Yes, initially the DL10 front end was just a normal DC76 handling terminals. Larsbrinkhoff (talk) 17:29, 12 April 2021 (CEST)

Where is CONFIG > now? I looked here: https://github.com/PDP-10/its/tree/master/src/system here but it doesn't seem to be there. Jnc (talk) 17:47, 12 April 2021 (CEST)

Many (all maybe?) versions are here: https://github.com/PDP-10/its-vault/tree/master/files/system/ Larsbrinkhoff (talk) 17:51, 12 April 2021 (CEST)

Ah, thanks.

It looks like it doesn't say what kind of tape drive it was; it was a TU77, or something. If we ever get good photos of the machine, we should be able to tell (also, if there was a second DF10).

I wonder why they did the whole DL10 thing just to get a few more serial lines. Maybe the DTE20 one was maxed out?

I see from the IOELEV/KLDCP source that the DTE20 -11 had a TU56; I vaguely remember that. Jnc (talk) 18:53, 12 April 2021 (CEST)

I only know of one photo of MC when at MIT, the one on the ITS page, but there's no tape drive. I saw MC in LCM storage, but I don't remember any peripherals. Larsbrinkhoff (talk) 18:59, 12 April 2021 (CEST)

Here is the layout of the machine, as best as I can remember it. The tape drive would not have been in the field of that image.

A couple of notes. I'm sure of the CPU, MF10's, Fabridrek Ampex, RP04's, Tape, Trident and DL10 locations. The RH10/DF10 were I think in that corner, but I'm not positive, and I don't remember which order. Ditto for the order of the DL10 and its PDP-11 (which I think was an 11/40). I think the TM10 was there somewhere, but I'm kind of guessing; I don't think it was next to the drive, but it might have been. Jnc (talk) 20:34, 12 April 2021 (CEST)

Ah, looking at that image, it's taken from just in front of the Ampex; you can see a corner of it. The RH10/DF10 are indeed in that corner; you can also see a bit of the RP04's behind them. The tape drive would be behind the camera; too bad we don't have a shot in that direction. Jnc (talk) 21:51, 12 April 2021 (CEST)

Very interesting, thanks! I suggest copying your comment to File_talk:Mit-mc.jpg. Larsbrinkhoff (talk) 20:58, 12 April 2021 (CEST)

Probably Talk:Incompatible Timesharing System would be better. I'll move the whole thread there. Jnc (talk) 21:51, 12 April 2021 (CEST)

Done. Jnc (talk) 14:05, 13 April 2021 (CEST)

I wish we could confirm what the new memory was; it was probably an Ampex ARM10, since other KL10's are reported to have had those (e.g. Guy S's machine); maybe some mail log file, or Moon's ITS changes log file, will mention it. Ah, there's an image of one the CHM has/had here (inside here), and that is indeed what was added to MC. (Well, I don't remember the inside well, but the outside matches.)

Every so often a module (CHM image here; they were huge, physically) would fail, and we'd have to take the machine out of 4-way interleave; same thing when an MF10 failed. Jnc (talk) 14:05, 13 April 2021 (CEST)

Also, in SYSDOC;POOR MC, MOON (who knew more about MC than any other human) refers to "MH10 C". So maybe my memory flaked (again, sigh) and they're actually MH10's, not MF10's. Jnc (talk) 17:20, 13 April 2021 (CEST)

But SYSDOC;KL10 FLKLOR refers to MF10's. So maybe they were upgraded at some point? Jnc (talk) 17:32, 13 April 2021 (CEST)

This email seems to indicate that something else (probably MF10's) were replaced by MH10's. Here's where it gets funny. That email is from ... {drum roll} ... ME!! I know I did a lot of drugs back then, but ... sheesh! Jnc (talk) 03:38, 4 June 2022 (CEST)

The timestamp of the KL10 FLKLOR file is from February 1976, so it seems likely the MF10's were part of the original configuration. Larsbrinkhoff (talk) 14:10, 4 June 2022 (CEST)

Since I was obviously there at the time, one would think I'd remember if the MF10's were replaced by MH10's, or merely upgraded. But I don't have the foggiest idea! I have no memory of this at all. (Which shows why, for historians, contemporary written evidence is the 'gold standard'!) I'd have to look at the MF10 and MH10 documentation, and see if such an upgrade was possible. I'd say that if so, there's a good chance that that is what happened, since moving 8 large, heavy MF10 cabinets out, and 8 more MH10's in, would have been a major upheaval that left an impression. But maybe not: this, along with the previous message, shows that I apparently did a fair amount of hardware work on MC (OK, that latter one was on the front end -11, and I knew -11's - although in this case I'm surprised that I had a spare H754 just lying around); in this latter case, I apparently decided on my own to attack the machine, and did the work by myself! But it was an -11 that was dead... Jnc (talk) 14:56, 5 June 2022 (CEST)

With the new images, I can confirm that the MF10 boxes were definitely replaced; the MF10's and MH10's had different DEC indicator panels, and those images show MF10's. Jnc (talk) 22:34, 4 February 2024 (CET)

There is a lot about failing Ampex ARM10 modules in Moon's files, so I would assume that confirms your recollection. I don't have any good, clear information about which ITS machines had what kind of memory, but Ampex is generally mentioned a lot. Given a lot of time, one could probably dig out an uneven coverage of part numbers, serial numbers, prices, dates, etc. Larsbrinkhoff (talk) 14:29, 13 April 2021 (CEST)

Yes, definitely an ARM-10L; having found an image of one, I recognize it. Jnc (talk) 08:40, 11 April 2022 (CEST)

Stephen Jones from the LCM called me last night; he's started exploring MC's remains (and the other stuff they got from Peter Löthberg - principally two CADR's), apparently with a view to getting it running, so in the near future we should have a confirmed list of everything on MC. It's still pretty complete, apparently (the CADR's are missing their keyboards). Jnc (talk) 13:24, 26 October 2022 (CEST)

DM/ML configuration

It shouldn't be too hard to figure out DM and ML's configuration; between my (and if needed, other people's) memory, CONFIG >, and similar sources.

Both had RP10's with a DF10. ML had RP03's (I think; definitely RP0x's), but I forget how many; DM I vaguely recall also had DEC disks. I recall that ML's had a slanted top front panel, which I think made them RP03's; those on DM had vertical panels, I think, which I think were RP02's. Per SYSDOC; POOR MC, both DM and ML (and also AI) had TU20 tape drives; one had a TM10A (no DF10, just an I/O bus connection), and one a TM10B (with DF10, but I don't know if the DF10 was shared with the RP10, or a separate one - a DF10 could be shared between several controllers); CONFIG > reveals that it was DM that had the DMA one.

For main memory, ML I think had 4 MD10s, @128KW each. DM I'm pretty sure had 2 MD10's, and a collection of other DEC early memory boxes (MA10's and/or MB10's) totalling 256KW. I can ask Jack Haverty (and old DM hacker) if he remembers the details. Tim Anderson might remember too. Both machines has Systems Concepts DM-10 mapping boxes, of course.

So that's a good start. Jnc (talk) 02:32, 14 April 2021 (CEST)

CONFIG says RP02 for both DM and ML, but ITS, SALV, and DSKDMP checks the type at run time and supports both RP02 and RP03.

I think the DF10 was shared between the disk and tape. This configuration is used with the KA10 emulator running ITS.

The KA10 DF10s had a special hack to make them work with 20-bit physical addresses.

I have collected some notes about the evolving memory configurations: https://github.com/PDP-10/its/issues/1585

Two diagrams of DM. Neither has a tape drive.

https://github.com/PDP-10/its/issues/181#issuecomment-483222759

https://github.com/PDP-10/its/issues/181#issuecomment-489417548

Larsbrinkhoff (talk) 08:21, 14 April 2021 (CEST)

[2] says DM had 3x RP02's and 3x RP03's; that sounds about right. CONFIG says ML had 7 drives, which sounds a bit high, but I suppose could (must?) be right. Ah, SHUT DOWN says it has 4 DEC disks and 4 (2 bays, 2-high) CalChomps (just like the ones on AI); that rings a bell, I recall the CalChomps now. (That memo also confirms DM's disks.)

DM definitely had a tape drive; CONFIG confirms that.

I think we're mostly done with these two now (although I have ignored the serial line controller(s) for now). Jnc (talk) 15:36, 14 April 2021 (CEST)

An archived email from GSB confirms that ML had at least one MD. Since I'm pretty sure it had 4 identical memory bays, that pretty much confirms it had 4xMD's. Jnc (talk) 14:52, 19 April 2021 (CEST)

POOR MC says that all the KA10 machines had TU20 magtape drives. There an image of one of them here, and it's definitely a TU20, not a TU10 - it has the horizontal vacuum columns of the TU20 (I'll do a page for it soon; the manuals are in Bitsavers), not the vertical ones of the TU10. Finally, this confirms that it's a TU20. So I think that's pretty definite. Jnc (talk) 23:16, 6 February 2024 (CET)

AI configuration

There was a long list of "fanciful hardware" (i.e. kludges) attached to the AI PDP-10.

For now, I'll point to Hardware memo 3 which says there was the 256K moby, plus a DEC 16K and an Ampex 16K. It seems one of those 16K memories would later go somewhere else.

When I was around, the smaller memories were powered off, and later flushed. The machine had 512KW: the 256KW Fabridrek moby, and another 256KW unit, about which I have been racking my brains all evening trying to remember, and can't quite. Later they both (I think) went away, and HIC added the memory box that used CADR memory boards.

One thing I'm curious about was how the memory bus worked, for more than 256KW (the vanilla KA memory bus only supported 256KW). I had assumed that the mapping box produced separate memory busses for each moby, so standard cables could be used, and the memories remain un-altered. The flaw in this theory is the DF10's; I suppose they could have been modified to use the same kludge, but...? Maybe ask RG? (He's know about the memory too.) Jnc (talk) 15:36, 14 April 2021 (CEST)

Note that AI didn't have a DF10; Systems Concepts' DC10 and DK10 had DMA access, and TM10A was IO bus only.

The DF10s were indeed modified: "MIT-ML and MIT-DM (also KA's) had 18-bit DF10's with a special kludge modification (disabled with a toggle switch for running DEC diagnostics) to take extra address bits from the complement of some high-order bits of the count. The complementation is so that with a small count (which is negative) you address the low-order memory, for compatibility." (David Moon, 1985, KS-ITS mailing list)

HW memo 2 documents the AI pager: https://github.com/larsbrinkhoff/its-archives/blob/master/ailab/ITS_Hardware_Memo_2.pdf

Larsbrinkhoff (talk) 16:05, 14 April 2021 (CEST)

I know about the functionality of the MIT DF10, which is actually already described here. I'm trying to work out the low-level details of how the cabling, etc, actually worked. According to e.g. the PDP-10 Interface Manua (DEC-10-HIFB-D), Chapter 7, the KA10 memory bus has only 18 bits of address. So if one has 512KW of memory ... how does one address it? I had a theory that maybe there was a separate 18-bit-address bus for each moby, but of course there are other possibilities. I'm not sure if the connector (pg. 124) has any spare pins; if so, they could be used - but would need an extra wire run from one end of the cable to the other. A better possibility is that a number of signals (A18-A22, A35) are present in positive and negative logic forms, and maybe one of the semi-duplicates could be disposed of, and the pin re-used for, say, A17.

I'm about to email Jack Haverty, to ask him about DM's memory configuration; I'll ask him, maybe he knows. If not, we can ask RG, he should know. Jnc (talk) 20:41, 14 April 2021 (CEST)

The same archived email from GSB (above) indicates that the second moby on AI was an ARM-10; all well and good but now we have to sort out all the ARM-10 variants (see Talk: Ampex ARM10 for more); there are at least 3 (including this one). Jnc (talk) 14:52, 19 April 2021 (CEST)

Still to do

I think we have most of this sorted out now. Left to do:

• Confirm whether AI-KS had a TU45 or TU78. EK-KS10-TM-PRE (and probably other places) say that 'DEC FS will not maintain systems that do not have a TU45', but . . . SYSTEM;TM78S DEFS2 exists. So maybe it had a TU78? On the gripping hand, NMTAPE 30 doesn't seem to use any symbols from there, it seems to only use things from TM03S DEFS5 (which doesn't seem to have anything drive-specific in it), so maybe it was a TU45 after all? Did the TM03 even support a TU78?
  • Now that I think about it, it was almost certainly a TU45. The TU78 uses a TM78 formatter, which plugs into the MASSBUS. But CONFIG 186 says that AI-KS had a TM03 ("DEFOPT TM03S==1"); and the TM03 also connects to the MASSBUS - i.e. one wouldn't have both. So if it didn't have a TM78 it could't have had a TU78. Q.E.D. (Although I suppose it could have had a TU77.) Jnc (talk) 17:25, 7 February 2024 (CET)
  • Maybe not a TU45: from KSHACK; KS-ITS MAIL1: "Date: Fri, 24 Oct 86 07:01 EDT From: Alan Bawden The drive is a TU45, which we have never tried with the ITS magtape code". I give up! Ah: "Date: Sat, 10 May 86 05:37:09 EDT From: Pandora B. Berman AI has a tu77". So maybe we have the answer. Jnc (talk) 05:05, 8 February 2024 (CET)
• Apparently none of the other KS10's had a tape drive, even MC-KS. This is plausible; they could have been backed up over the Chaosnet. It would be good to confirm that, though.
• Did MD-KS really have an RM80 Disk Drive, and nothing else? The thing is, that's a non-removable-media disk, so . . . how did the bits get into the machine, to start with? I guess it would have been feasible to cable that drive into another machine to load bits onto it, though. (MC-KS and ML-KS both had RP06's, so they could have used packs written on AI-KS.)
• It would be nice to know what the other moby on DM (not the two MD10's was); I vaguely recollect it being a mashup of early DEC memory boxes (MA10's, MB10's). That will probably be hard to figure out.
• Serial line interfaces on all the machines (other than MC).
• Did the TM10B on DM have its own DF10, or did it share the RP10's?

I guess that's all I have for the moment. Jnc (talk) 03:31, 7 February 2024 (CET)

A good source

See the Arpanet Resources Handbook:

• AI: pg. 355-370
• DM: pg. 373-389
• MC: pg. 391-398
• ML: pg. 399-404

Look in the 'Hardware' section. (The other sections are interesting too.) Don't take that all too literally; e.g. the ML entry says it has "6 x RP02" disks, which was never correct while I was there; the RP02 has a vertical front face, and the RP03 a slanted one (see the images I am about to add), and I'm fairly sure ML had RP03's - and only 4 of them. (ISTR that the salvager or something will confirm the sizes, which will confirm.) SHUT DOWN confirms my memory that AI had 8 Calcomp 215s (also coming soon), not 6. And DM never had an RK05. I'm not sure about AI having two TU20s; it's possible. But that list is generally correct. Jnc (talk) 17:43, 10 February 2024 (CET)

Fabridreck

Old Tech Sq jargon! (I didn't create it; I just remember it being used.) I assume that was because, compared to later memories (it was really early, having been attached to the -6), it was temperamental, and more flaky. (Although obviously not so temperamental and flaky that it was unusable in a large time-sharing system.)

I was wondering, when I first saw those speed numbers in the ARH, if, based on the speed difference between it and the Ampex, if the Ampex was assigned to a low absolute address, so the OS was housed in it, leaving the Fabritek to be used entirely for housing user code? 'Obviously', if one has two speeds of memory (and the speed difference was really large, which was why I thought it worth noting), it would generally (I am guessing) be better to have the OS running out of the faster memory. (On a general-purpose time-sharing system, with lots of simultaneous users; on a system primarily used for big number-crunching applications, not so much.)

If so (on the location assignment guess), a side-benefit would be that the 'flakier' memory would hold user processes, and the system is more likely to be able to deal with memory errors in user processes than those in the OS. (I don't recall how ITS handled the two cases, but in UNIX if it was user memory that got the error it just kills the process:

               printf("mem error");
               if (cputype == 70)
                       printf(":  %o %o", MEMORY->r[1], MEMORY->r[0]);
               if (dev & USER) {
                       i = SIGBUS;
                       break;
                       }
               panic("kernel mode mem fault");

and continues on its merry way, while a memory error in the OS kills the system.)

I don't know if the Fabritek really was more temperamental and flaky than the Ampex (i.e. there was data showing that it was), or if the hackers just assumed it was. Jnc (talk) 16:39, 11 February 2024 (CET)

Huh, I have actually never seen "fabridreck" before. I understand it took a year or so from delivery for the full 256K to be working (somewhat?) reliably, so it certainly seen to have been flaky from the beginning. I don't know which memories went where in the address space.

I have seen plenty of other nicknames like memowreck, dissapoint, etc. So I can see it was a tradition. Maybe these should be collected in a separate article. Larsbrinkhoff (talk) 07:56, 12 February 2024 (CET)

Go for it! Jnc (talk) 13:18, 12 February 2024 (CET)

Plasma

I find it confusing, but I believe there were two "Plasma" groups/departments, correct? One in Building 38, sometimes referred to as the Plasma Group (and Plasma Physics?). Another in NW16, the Plasma Fusion Center. The one in Building 38 had the Gould printer and a TV system with Grinnell frame buffers; maybe it was called the R11/RTTY11/CHSGTV. The one in NW16 also had a TV system; maybe KTV.