"I've spent the night reading Tandem tech reports and manuals."
Wow. Great, detailed write-up from a night. I've been waiting for one like this. You should look at the links to the hardware in the Tandem or NonStop wikipedia articles, though, as the tradeoffs made and techniques used were sometimes interesting.
" instead, it has to do with process naming, sequence numbers, and copying around memory space images between process pairs running on separate "CPUs", so that if one of them fails, the other one can pick up from the last checkpoint."
A bit reminiscient of the cache-coherence & other memory protocols in NUMA machines. Just for availability as much as scaling.
" they had a 300ms heartbeat from the beginning."
Costs money but works. :)
" I can find no trace of Alpha, PA-RISC, or i386. Also, I don't think the issue there was acquisitions;"
It's in the Wikipedia article along with some of those hardware details I referenced above:
They got acquired by DEC. Half-way through Itanium port, idiots in management forced them to switch it to Alpha. Then Compaq killed Alpha itself. Later, Intel started killing Itanium. NonStop ran into bad management and bad luck. I see no PA-RISC or vanilla x86. Actually, Itanium was a joint venture between Intel and HP to replace x86 and PA-RISC with something having benefits of both plus experimental stuff. Porting it to PA-RISC instead of Itanium would be downgrade.
"They claimed from the beginning that their hardware cost was comparable to competitors, but maybe they were lying."
IBM's prices are so inlfated that many mainframes competed on price. I read before Fujitsu's were 50-75% as cheap. Far as NonStop, it's around $120,000-$2.7million. Quotes I heard on IBM mainframes were much higher esp on high end.
Heartbeats, to a first approximation, don't cost money. You could do an entirely reasonable heartbeat using parallel ports (like Princeton PAPERS, but for failover rather than HPC), sound cards, or a couple of GPIO pins. What heartbeats cost is the ability to use off-the-shelf networking hardware and software, because it's all best-effort — nothing based on Ethernet tries to guarantee delivery, much less latency.
I guess it makes sense that if they never shipped an Alpha product, their tech reports and manuals would never mention it. I guess John was right about that!
I don't think Tandem was comparing their prices to IBM prices in the early years, though I could be wrong; I think they were comparing them to the prices of the Seven Dwarves.
"Heartbeats, to a first approximation, don't cost money. "
Over a WAN? The low-latency connections were an important part of their WAN high-availability per Animats. Leased lines here cost more than a house payment.
"I think they were comparing them to the prices of the Seven Dwarves."
Well, they were called the Seven Dwarves because IBM dominated the market. So, Tandem was taking on IBM and all of the rest. They had to be price-competitive with Seven Dwarves while looking way better than IBM, esp price/performance. Them getting SQL to scale linearly in NonStop SQL was one of their differentiators IIRC.
Wow. Great, detailed write-up from a night. I've been waiting for one like this. You should look at the links to the hardware in the Tandem or NonStop wikipedia articles, though, as the tradeoffs made and techniques used were sometimes interesting.
" instead, it has to do with process naming, sequence numbers, and copying around memory space images between process pairs running on separate "CPUs", so that if one of them fails, the other one can pick up from the last checkpoint."
A bit reminiscient of the cache-coherence & other memory protocols in NUMA machines. Just for availability as much as scaling.
" they had a 300ms heartbeat from the beginning."
Costs money but works. :)
" I can find no trace of Alpha, PA-RISC, or i386. Also, I don't think the issue there was acquisitions;"
It's in the Wikipedia article along with some of those hardware details I referenced above:
https://en.wikipedia.org/wiki/Tandem_Computers
They got acquired by DEC. Half-way through Itanium port, idiots in management forced them to switch it to Alpha. Then Compaq killed Alpha itself. Later, Intel started killing Itanium. NonStop ran into bad management and bad luck. I see no PA-RISC or vanilla x86. Actually, Itanium was a joint venture between Intel and HP to replace x86 and PA-RISC with something having benefits of both plus experimental stuff. Porting it to PA-RISC instead of Itanium would be downgrade.
"They claimed from the beginning that their hardware cost was comparable to competitors, but maybe they were lying."
IBM's prices are so inlfated that many mainframes competed on price. I read before Fujitsu's were 50-75% as cheap. Far as NonStop, it's around $120,000-$2.7million. Quotes I heard on IBM mainframes were much higher esp on high end.
http://www.fundinguniverse.com/company-histories/tandem-comp...