Slashdot Mirror
Server Runs Continuously For 24 Years (computerworld.com)
In 1993 a Stratus server was booted up by an IT application architect -- and it's still running.
An anonymous reader writes:
"It never shut down on its own because of a fault it couldn't handle," says Phil Hogan, who's maintained the server for 24 years. That's what happens when you include redundant components. "Over the years, disk drives, power supplies and some other components have been replaced but Hogan estimates that close to 80% of the system is original," according to Computerworld.
There's no service contract -- he maintains the server with third-party vendors rather than going back to the manufacturer, who says they "probably" still have the parts in stock. And while he believes the server's proprietary operating system hasn't been updated in 15 years, Hogan says "It's been extremely stable."
The server will finally be retired in April, and while the manufacturer says there's some more Stratus servers that have been running for at least 20 years -- this one seems to be the oldest.
There's no service contract -- he maintains the server with third-party vendors rather than going back to the manufacturer, who says they "probably" still have the parts in stock. And while he believes the server's proprietary operating system hasn't been updated in 15 years, Hogan says "It's been extremely stable."
The server will finally be retired in April, and while the manufacturer says there's some more Stratus servers that have been running for at least 20 years -- this one seems to be the oldest.
Agreed. Based on one of the embedded articles reference in the original the server had run for at least 4 years continuously, which I still find impressive.
http://www.computerworld.com/article/2550661/data-center/this-server-outlasts-two-presidents.html
Stratus has proprietary redundant *everything* on their machines, and runs in lockstep; they literally have two of everything in there... two motherboards, two cpus, two sets of RAM, etc. If anything weird happens on one side, they fail over to the other motherboard running in lockstep on the other blade in the chassis. Combine that with running an extremely conservative set of drivers that are known stable, and you can get six nines out of the thing. Stratus is typically used for credit card processing and banking applications where it's not ever acceptable to have a machine down for the time it takes to reboot. Really, really, really expensive though. You wouldn't want to use one of these for anything normal.
Since this machine is running VOS, and from the '93 time frame it's either an X/AR with i860's or a Continuum with PA-RISC. I'll spitball and say it's a Continuum.
These machines are not like desktops. The hardware and software is extremely tightly coupled. Multiyear uptimes are not uncommon on Stratus VOS machines.
Full disclosure, I'm a former Stratus Employee.
Yes Francis, the world has gone crazy.
Why do you say that's less impressive than running 24 years continuously? Any non-trivial application requires servicing eventually.
And how will you even be able to tell if that is the case?
In a virtualization environment; I have servers with 7 year uptimes. Of course, they have occasionally been vMotioned between hosts -- in some cases, servers have been checkpointed, Suspended for a few hours, then resumed in another datacenter without any operating system reboot, so if you go by OS uptime they've been up for 10 years.
Sometimes a server application can become stalled or break, So it's not provided continuous service, but there's no visible indication on the server, no administrative indication in the log, etc.
I used to work at Stratus. Most components were duplicated (disks, IO boards, power supplies). Everything was hot swap. CPU's were duplicated *pairs*. Each pair ran in lockstep, instruction by instruction, and results were compared within a pair. If one pair pf CPUs disagreed (between the CPUs in that pair), but the other pair agreed (between the CPUs in that second pair), the first pair was taken off line and the second pair continued processing.
Each pair of CPUs (and their associated memory) were on a separate board. The faulty board would light a red light, and the admin could pull that faulty board with the system running. A replacement board could be installed, again with the system running transactions. When the new board was installed, a process would start of synchronizing its memory with the content of the memory of the good board. This was done with the system running (processing bank transactions, for example), so the bus bandwidth between the boards had to be fast enough to be able to handle the rate at which the memory contents on the good board was changing. At a certain point the memory state of the two boards would be in sync and the new board would start processing, again in lockstep with the board that had been running all along. The repair process was so easy that one engineering director there had what he called the "mom test" - he have his mom come in and see if she could fix a system that had been forced to throw a fault. Red light? Pull that board out and put a new one in. New board's lights went red-yellow-green (as memory was brought into sync), and you're running fault-tolerant again. Easy peasy. (When the boards failed, they'd "phone home" and Stratus tech support would know there was a fault often before the customer did. They'd ship a replacement part overnight to the site with the bad part. Anyone who worked at Stratus in those days knew the story of the FedEx driver doing a repair for a customer.)
OS upgrades were done by un-synchronizing the OS drives and upgrading one at a time. One would be taken off line for an OS upgrade. The system would have to go through a restart to run on the upgraded OS drive (that would be done in a planned maintenance window), but once it was running, the second OS drive would be mirrored to the running OS drive, at which point the disks were redundant again. The unplanned downtime was zero, the planned downtime was minimized.
It became a challenge for Stratus when CPUs became nondeterministic (instructions wouldn't necessarily process in exactly the same order, making lock-step processing a real problem). At least one CPU architecture transition was driven by that issue. And clearing the heat from 4 cpus in a small space was a thermal challenge. But they were reliable beasts, expensive enough only to be used for workloads involving real money (e.g., financial transactions). Back when the founder (Bill Foster) was still CEO, it was a great place to work. When he left and the MBA's moved in, the company got sold to Ascend Communications, then Ascend got bought by Lucent, the non-telecom part of the business was spun off (as Stratus Technologies), and yes they are still in business.