VOGONS


First post, by donkom

User metadata
Rank Newbie
Rank
Newbie

This is something that has eluded me for a very long time. I'm new to this forum, so thanks for entertaining these thoughts!

Up until today, I didn't even have a photo of the outside of these machines. Found a great article on CPUshack that showcased this view, and specs of such a beast: http://www.cpushack.com/wp-content/uploads/20 … 028_xr6_big.jpg (full article here: http://www.cpushack.com/2019/01/12/mini-mainf … rver-from-1997/ )

I own an Aquanta HS/6 which has been discussed on this forum before. It's a beauty and a beast. I know that Pentium Pro CPUs were designed for a maximum of 4-way connections without any hardware "glue", and I believe the Aquanta HS/6 (aka ALR 6x6) used something like this to get to six CPUs. Simple to imagine that the fourth processor in a cluster wasn't actually a CPU but a relay to another cluster, where every CPU logically sees four, but one of the four is actually another three. I cannot understand how this would scale farther and still be compatible with standard operating systems like Windows NT 4.0. The ASCI Red supercomputer was built using these chips, breaking the teraflop barrier, but supercomputers always beat to their own drum.

So, I'm just immensely curious: does anyone have any experience with an Aquanta XR/6, even seen one, or know of any resource for photos of the inside of such a beast? It's a dream to one day own one, if any have been saved from the scrap heap. At the very least, this is such an interesting hardware story that has never been told and I'd love to know more.

Reply 1 of 5, by Swiego

User metadata
Rank Member
Rank
Member

At the very start of my career I worked on an Oracle database running on a Sequent server running if I recall 6x 80486s. We decommissioned it for Y2K and I remember playing wolf3d on it before it was to be... well I’m not sure what it’s fate was. I do have fond memories of that system and thus can appreciate your fondness for the Aquanta. Wish I could offer more help.

Reply 2 of 5, by PC Hoarder Patrol

User metadata
Rank Oldbie
Rank
Oldbie

There's still a few partial pages available on the Unisys Wayback archive

Specs - https://web.archive.org/web/20000707221647/ht … ervers-xr6.html

Tech Brief 1 (Intel MPS / System Overview) - https://web.archive.org/web/20001006223456/ht … ary/wp-xr6.html

Tech Brief 2 (Ten-Way System Architecture) - https://web.archive.org/web/20001006195316/ht … /wp-tenway.html

Reply 3 of 5, by dionb

User metadata
Rank l33t
Rank
l33t
donkom wrote on 2020-08-04, 01:11:

[...] I cannot understand how this would scale farther

Just hypothesizing it's conceptually quite simple:

PPro's internal logic supports clusters of 4, so you do 3 CPUs and the glue switch, as you describe with the /6. However instead of the glue just connecting two of those groups of 3 CPUs, the glue itself is connected in a 4-way group, so each glue switch is connected downstream to three PPro CPUs and upstream to three other glue switches, giving you a total of 12 CPUs.

Now, to actually implement that is quite another matter, but the kind of people who could do that certainly worked at Unisys at the time.

and still be compatible with standard operating systems like Windows NT 4.0. The ASCI Red supercomputer was built using these chips, breaking the teraflop barrier, but supercomputers always beat to their own drum.

Regular MS NT 4.0 HALs went up to 8 CPUs max, but vendors of this sort of system wrote custom HALs, Compaq wrote a 32 CPU one at one point. That said, it's unlikely that people using this sort of monster in PPro days would have run Windows on it. ASCI Red ran different OS like Teraflops OS and Cougar, custom made for the hardware. Linux was also ported to work on ASCI Red, but its performance was significantly lower than Cougar.

Reply 4 of 5, by chinny22

User metadata
Rank l33t
Rank
l33t

Very interesting!
from a quick read of the Tech Brief 1 linked above it looks like the HAL somewhat emulates a cluster to get around the CPU limit.
I guess that would be the sensible way to go with the amount of RAM and CPU's been talked about here far surpassing typical servers of the time .
but yeh, thanks for the information. Really like this old enterprise stuff although sadly most of it has long gone forever

Reply 5 of 5, by donkom

User metadata
Rank Newbie
Rank
Newbie

Swiego, thanks very much for the links!

In one of the tech briefs, we get a confusing clue: "The Aquanta XR/6 system uses processor boards that contain two Pentium Pro processors and an L3 cache module. The system can support a maximum of five processor boards (or 10 processors)."

This was for their initial 10-way system, later upgraded to 12-way. That would make sense to have four triple-CPU boards (as dionb describes). But there are five boards? Why didn't they go to town and make a 15-way system? I also found it interesting that the maximum amount of memory was cut down to 4GB when in a 12-way configuration. It also seems like the original 10-way system was not built around 1M Pentium Pros, but the 512K versions.

I'm sure there was so much special glue, both in hardware and software, to make this all work. I'm sure that my attempts to run Linux on my 6x6 system could be improved with custom code that no longer exists in modern software. I recall something about the ASCI Red having different operating systems for different segments of the computer. Would have loved to dumpster dive after the upgraded it to Pentium II OverDrives or when it was decommissioned.