The Design of the Connection Machine (1994) | Hacker News Reader

The Design of the Connection Machine (1994)

mission-base.com

17 pointsrobin_reala6 comments

I've been fascinated with this machine for ages... I can't believe 20 years have passed so quickly.

From time to time I read about R&D projects which remind of it.

When Adapteva started talking up their Epiphany Chip & Parallella board, I had hopes that it would be possible to construct a machine which was vaguely reminiscent. That hasn't really come to pass. They've really struggled to get the first generation designs out and they've had to step back from older production schedules. But they haven't completely failed either; you can actually buy boards with 16 core devices now. http://www.adapteva.com/

The Micron Automata Processor Project is also vaguely similar and interesting in its own right. http://www.micron.com/about/innovations/automata-processing

There’s also Intel’s Xeon Phi boards with 72 cores. I suppose the more interesting thing though about Connection Machines was the routing though, something that Intel solves with a reasonably standard ring bus.

Yep and right now there is a promotion for the Intel Xeon Phi 31S1P. This is a PCIe board with 8Gig of RAM and passive cooling. So it demands a motherboard which supports it and some sort of add-on cooling solution.

I've had a very long term personal pet project of social simulation. Multi-Agent sims are really well suited for systems with very many independent cores (threads); that's why I was interested in the Epiphany. Seeing the Phi selling for something like $200 is really, really tempting. However, I hesitate because not only is the most of development that's gone on for the Phi so far really different than what I've been doing, the next version of the Xeon Phi will be self hosting (so it will be in a regular processor socket in a regular server motherboard). So I expect the prevalent software architecture to change again and the PCIe based versions will become a sort of evolutionary dead end.

Still... I have a Haswell based development box which would support the Phi so it's really tempting! My secret hope is they begin to surface on eBay from folks who impulsively buy them only to discover they don't have a system which actually supports it and whatever they thought they were going to do isn't worth buying additional hardware.

"the more interesting thing though about Connection Machines was the routing"

Indeed. This part of the article goes into it, including Feynman's participation: http://www.mission-base.com/tamiko/theory/cm_txts/di-ch2.htm...

While I worked at the Naval Research Lab they acquired a series of these machines as they came out. They were very fast for "embarrassingly parallel" problems, and problems where computation dominated communication. I published a few papers on shocks and detonations in crystals from simulations using these machines. They used a language called "C*" that was C with parallel extensions specifically for the CM.

Also, the CM-5 moved away from the hypercube topology of the CM-2 to a fat tree.

Danny Hillis' PhD thesis, "The Connection Machine": http://dspace.mit.edu/handle/1721.1/14719

A great read.

Reply on news.ycombinator.com