9 Dec
2013
9 Dec
'13
10:07 p.m.
Hi all,
I am attempting to get Helios running on a 64-core T805-based Transputer
array. This is a one-of-a-kind system with a custom case. It uses a
single-board 486DX with 8MB of RAM on an ISA card for the host system. The
PC front end runs Windows 3.1 or MS-DOS 6. There are eight T805s on eight
custom cards. It doesn't use TRAMs, but I have a feeling these are wired
like such on the board. Pictures of the system can be found here:
http://imgur.com/a/fCA3C
I've been getting all of my Helios editions from
http://www.classiccmp.org/transputer/helios.htm. I first tried version 1.1,
and after copying the contents over to the hard drive from a single floppy
disk, running the server results in an error:
ns : Incompatible version of Rmap 6e627573
ns : error is fatal, exiting.
I finally got 1.31 working on the system by downloading another DOS
executable of the server and playing with some configurations. I also used
ispy and mtest to see the structure of the system, which uses several C004
cross links. Unfortunately, ispy only reports a single T805-25 processor
connected to the host at address 0x150.
Does anyone have any ideas as to why the other 63 T805s and several C004s
aren't showing up?
Thanks in advance,
Kyle
12 Dec
12 Dec
9:13 p.m.
Hi Kyle, not seen a system like that before - looks very nice!
No idea on the Helios boot I'm afraid, but do please post back to the
group if you make progress because I'm really keen to try and get it
running on my 340 node T425 system when I get some time :)
James
On 09/12/2013 22:07, Kyle Owen wrote:
Hi all,
I am attempting to get Helios running on a 64-core T805-based Transputer
array. This is a one-of-a-kind system with a custom case. It uses a
single-board 486DX with 8MB of RAM on an ISA card for the host system. The
PC front end runs Windows 3.1 or MS-DOS 6. There are eight T805s on eight
custom cards. It doesn't use TRAMs, but I have a feeling these are wired
like such on the board. Pictures of the system can be found here:
http://imgur.com/a/fCA3C
I've been getting all of my Helios editions from
http://www.classiccmp.org/transputer/helios.htm. I first tried version 1.1,
and after copying the contents over to the hard drive from a single floppy
disk, running the server results in an error:
ns : Incompatible version of Rmap 6e627573
ns : error is fatal, exiting.
I finally got 1.31 working on the system by downloading another DOS
executable of the server and playing with some configurations. I also used
ispy and mtest to see the structure of the system, which uses several C004
cross links. Unfortunately, ispy only reports a single T805-25 processor
connected to the host at address 0x150.
Does anyone have any ideas as to why the other 63 T805s and several C004s
aren't showing up?
Thanks in advance,
Kyle
13 Dec
13 Dec
3:22 a.m.
On 12/12/13 4:13 PM, James Wilson wrote:
Hi Kyle, not seen a system like that before - looks
very nice!
No idea on the Helios boot I'm afraid, but do please post back to the
group if you make progress because I'm really keen to try and get it
running on my 340 node T425 system when I get some time :)
WOW. Wow. I would love to hear more, when you fire it up. Can you blog
it? Post pr0^H^H^H pictures??
I only ever ran two of the buggers, T800s, and they were fantastic for
both integer and floating point.
--Toby
James
On 09/12/2013 22:07, Kyle Owen wrote:
Hi all,
I am attempting to get Helios running on a 64-core T805-based Transputer
array. This is a one-of-a-kind system with a custom case. It uses a
single-board 486DX with 8MB of RAM on an ISA card for the host system.
The
PC front end runs Windows 3.1 or MS-DOS 6. There are eight T805s on eight
custom cards. It doesn't use TRAMs, but I have a feeling these are wired
like such on the board. Pictures of the system can be found here:
http://imgur.com/a/fCA3C
I've been getting all of my Helios editions from
http://www.classiccmp.org/transputer/helios.htm. I first tried version
1.1,
and after copying the contents over to the hard drive from a single
floppy
disk, running the server results in an error:
ns : Incompatible version of Rmap 6e627573
ns : error is fatal, exiting.
I finally got 1.31 working on the system by downloading another DOS
executable of the server and playing with some configurations. I also
used
ispy and mtest to see the structure of the system, which uses several
C004
cross links. Unfortunately, ispy only reports a single T805-25 processor
connected to the host at address 0x150.
Does anyone have any ideas as to why the other 63 T805s and several C004s
aren't showing up?
Thanks in advance,
Kyle
9:05 p.m.
On 13/12/2013 03:22, Toby Thain wrote:
WOW. Wow. I would love to hear more, when you fire it up. Can you blog
it? Post pr0^H^H^H pictures??
I only ever ran two of the buggers, T800s, and they were fantastic for
both integer and floating point.
--Toby
Hi Toby, yeh it's a system I've had for many years and 2 house moves. I
do sometimes find enough time to power it up and start poking about. It
was originally a data mining system built by Whitecross. Organised as 6
T425s per card, each with 16MB ECC RAM. There are 19 cards per sub rack,
and fibre interconnects between each sub-rack. I have 3 "compute" racks,
some storage racks & 2 system controllers (it was originally built as 2
identical systems for redundancy) and 3 48V PSUs - each card has voltage
regulators for the 5V supply. Sadly I couldn't take the original
cabinets so it's housed in a standard open 19" rack now. I'm also very
lucky (Whitecross got in touch when they heard I rescued their largest
system!) to have a pair of specially configured Sun workstations with
custom sbus/fibre adapters (connected via a TTM50 SCSI interface) and
software to run diagnostics on the hardware. Here's some old pics and
documents:
http://www.machineroom.plus.com/
Picked up at a scappers for ?1.20 - they had no clue what it was ;)
Cheers!
James
3:50 a.m.
On Thu, Dec 12, 2013 at 3:13 PM, James Wilson <james at machineroom.info>wrote:
Hi Kyle, not seen a system like that before - looks
very nice!
No idea on the Helios boot I'm afraid, but do please post back to the
group if you make progress because I'm really keen to try and get it
running on my 340 node T425 system when I get some time :)
James
That sounds like a most impressive array, James! It would be awesome if you
could get some pictures of your system. I think it's fair to say that we
all share Toby's sentiments. And what was its purpose?
I never did get any further on the system this week. And unfortunately, the
next time I'll be able to play with it will be next year. I just don't know
what would cause only one T805 and no C004s to show up with ispy.
The best guide I've found to get Helios initially running is
http://www.geekdot.com/helios.html. Unfortunately, there doesn't seem to be
any good, general diagnostics for investigating missing nodes. Also, it's a
bummer that getting an X server running requires an Inmos video card, from
what my research shows. I may actually have a VGA card with an INMOS chip
on it, but it may just be some sort of dual-port RAM. I'll need to dig that
out and do some investigating.
Kyle
10:21 p.m.
> I am attempting to get Helios running on a
64-core T805-based Transputer
> array. This is a one-of-a-kind system with a custom case. It uses a
> single-board 486DX with 8MB of RAM on an ISA card for the host system. The
> PC front end runs Windows 3.1 or MS-DOS 6. There are eight T805s on eight
> custom cards. It doesn't use TRAMs, but I have a feeling these are wired
> like such on the board. Pictures of the system can be found here:
The circuitry probably is very similar to the TRAMs. The external
connectiosn ot a TRAM are the 4 transputer links, the reset/error/analyse
signals, a 5MHz clock inotu and power. No other trasnputer signals
(AFAIK) are brought out.
The T805 IIRC has an intenral DRAM controller. It will generate the RAS/,
CAS/ and address multiplex signals. All you have to add to link up DRAM
is the address multiplexer and maybe an address latch. Most 'compute'
TRAMs (That is, ones without special I/O devices like SCSI, HPIB or
ethernet) are jsut that.
It is very likely that;s what is done on your boards too.
One many TRAM motherboards the links go to C004 link switches so you can
configure the network as you like. It sounds like you have those too.
And given that, there aren't many other ways it could be wired.
-tony
27 Jan
27 Jan
3:22 a.m.
On Thu, Dec 12, 2013 at 3:13 PM, James Wilson <james at machineroom.info>wrote:
Hi Kyle, not seen a system like that before - looks
very nice!
No idea on the Helios boot I'm afraid, but do please post back to the
group if you make progress because I'm really keen to try and get it
running on my 340 node T425 system when I get some time :)
James
Hope I'm not resuscitating too old of a thread here, but I was able to get
more Transputers linking together. My 64 node array finally has most of its
nodes connected. Right now, the "CHECKOUT" program (a network worm utility)
is reporting that 63 nodes are interconnected. I'll be parsing this out to
form a network map compatible for Helios soon.
Here's what I've figured out: the entire network is configured (via the
C004 crossbars, presumably) using a "SETCROSS.EXE" utility, custom written
I believe for this system. I have the source code (written in C), but
thanks to poor documentation, I still can't figure out how the links are
numbered. Everything is quite cryptic to me.
Here is the data file that I've used to successfully configure the majority
(all but one T805, seemingly) of the Transputer array:
http://pastebin.com/1CUQDET0
This is the output from the "CHECKOUT" utility after running
"SETCROSS":
http://pastebin.com/XqHmuHvt
Here's the source for "SETCROSS": http://pastebin.com/NrY1mbPq
There's also a crossbar configuration generator, which takes in the size of
the matrix and outputs the data file for "SETCROSS" to read. It doesn't
work correctly; running "SETCROSS" results in only four processors
connected according to the worm. Here's the source anyways:
http://pastebin.com/wmtKzCSw and the header: http://pastebin.com/8yX8jvYd
This is the non-working output generated by the program:
http://pastebin.com/Ep7gpKt0
Lastly, I found a test program that worked with four of the T805s in the
system to generate some basic benchmarks, as well as draw some pretty
fractals. Here's an album: http://imgur.com/a/fCA3C
If anyone cares to shed some light onto how they think "SETCROSS" parses
the numbers into actual links, I'd be very grateful.
Thanks,
Kyle
28 Jan
28 Jan
9:41 p.m.
On 27/01/2014 03:22, Kyle Owen wrote:
On Thu, Dec 12, 2013 at 3:13 PM, James Wilson
<james at machineroom.info>wrote:
Hi Kyle, are you able to re-build the setcross program? I have the C
toolset somewhere if that would help. First step I'd take is to chuck
some console logging statements in that code, probably using good 'ol
printf(). The number of return paths from each function is truly scary,
so I'd tackle that first to get an idea of where the code is failing, if
indeed it is. Also image the disk in case you lose the other tools!
I must admit to not being familar with the crossbar switches, do you
have the data sheet? you should be able to find it on Rams page. Knowing
how the magic port addresses in the code relate to the hardware would be
a good start, you should be able to work backward from there. In other
words, use the crossbar data sheet as a known good starting point, work
backwards from there to understand the code and the data input format.
Litter with printf() if you get stuck :)
Cheers
James
Hi Kyle, not seen a system like that before -
looks very nice!
No idea on the Helios boot I'm afraid, but do please post back to the
group if you make progress because I'm really keen to try and get it
running on my 340 node T425 system when I get some time :)
James
Hope I'm not resuscitating too old of a thread here, but I was able to get
more Transputers linking together. My 64 node array finally has most of its
nodes connected. Right now, the "CHECKOUT" program (a network worm utility)
is reporting that 63 nodes are interconnected. I'll be parsing this out to
form a network map compatible for Helios soon.
Here's what I've figured out: the entire network is configured (via the
C004 crossbars, presumably) using a "SETCROSS.EXE" utility, custom written
I believe for this system. I have the source code (written in C), but
thanks to poor documentation, I still can't figure out how the links are
numbered. Everything is quite cryptic to me.
Here is the data file that I've used to successfully configure the majority
(all but one T805, seemingly) of the Transputer array:
http://pastebin.com/1CUQDET0
This is the output from the "CHECKOUT" utility after running
"SETCROSS":
http://pastebin.com/XqHmuHvt
Here's the source for "SETCROSS": http://pastebin.com/NrY1mbPq
There's also a crossbar configuration generator, which takes in the size of
the matrix and outputs the data file for "SETCROSS" to read. It doesn't
work correctly; running "SETCROSS" results in only four processors
connected according to the worm. Here's the source anyways:
http://pastebin.com/wmtKzCSw and the header: http://pastebin.com/8yX8jvYd
This is the non-working output generated by the program:
http://pastebin.com/Ep7gpKt0
Lastly, I found a test program that worked with four of the T805s in the
system to generate some basic benchmarks, as well as draw some pretty
fractals. Here's an album: http://imgur.com/a/fCA3C
If anyone cares to shed some light onto how they think "SETCROSS" parses
the numbers into actual links, I'd be very grateful.
Thanks,
Kyle
29 Jan
29 Jan
3:33 a.m.
On Tue, Jan 28, 2014 at 3:41 PM, James Wilson <james at machineroom.info>wrote:
Hi Kyle, are you able to re-build the setcross program? I have the C
toolset somewhere if that would help. First step I'd take is to chuck some
console logging statements in that code, probably using good 'ol printf().
The number of return paths from each function is truly scary, so I'd tackle
that first to get an idea of where the code is failing, if indeed it is.
Also image the disk in case you lose the other tools!
I must admit to not being familar with the crossbar switches, do you have
the data sheet? you should be able to find it on Rams page. Knowing how the
magic port addresses in the code relate to the hardware would be a good
start, you should be able to work backward from there. In other words, use
the crossbar data sheet as a known good starting point, work backwards from
there to understand the code and the data input format. Litter with
printf() if you get stuck :)
Yes, I do in fact have the tools to rebuild it, but there are already
plenty of executables on the disk. This machine was shared among many
graduate students, and it seems as though every directory has a copy of
SETCROSS.EXE! The program itself definitely works, and there's not much to
it. What I seem to have an issue with is how the array is architecturally
laid out. The DATA.DAT file that SETCROSS reads is just a bunch of numbers,
supposedly identifying a crossbar switch and links. But those numbers
themselves seem to have very little correlation with anything.
Here's the datasheet for a C004:
http://bitsavers.trailing-edge.com/pdf/inmos/transputer/C004_Programmable_L…
Unfortunately, there's more to the system than just a bunch of T805s,
C004s, RAM, and a host computer. There are also some custom Altera CPLDs on
each board. I'm unsure of their functionality, but I assume they are
helping the crossbar switches do their work.
Kyle
3766
days inactive
3817
days old
8 comments
4 participants
participants (4)
-
ard＠p850ug1.demon.co.uk -
james＠machineroom.info -
kylevowen＠gmail.com -
toby＠telegraphics.com.au