@Jplonie@jalefkowit
You're not joking. I used to work for a company that made redundant systems. (It wasn't Tandem.) They were pompously advertised as having infinite MTBF, because marketing people are like that and they plug their ears whenever Engineering talks to them.
What actually happened was that a bug took out node 1, and so node 2 took over node 1's workload and immediately crashed in the exactly the same way. Unsurprisingly, adding more nodes (beyond 2) didn't make the system any more reliable: they just produced a bigger and less predictable boot loop.
Not every failure would cascade across the cluster like this, but enough bugs did to make the infinite-MTBF claim laughable.
Edit: typo.
@Jplonie@jalefkowit
You're not joking. I used to work for a company that made redundant systems. (It wasn't Tandem.) They were pompously advertised as having infinite MTBF, because marketing people are like that and they plug their ears whenever Engineering talks to them.
What actually happened was that a bug took out node 1, and so node 2 took over node 1's workload and immediately crashed in the exactly the same way. Unsurprisingly, adding more nodes (beyond 2) didn't make the system any more...
Gregory's Server
@Jplonie @jalefkowit
You're not joking. I used to work for a company that made redundant systems. (It wasn't Tandem.) They were pompously advertised as having infinite MTBF, because marketing people are like that and they plug their ears whenever Engineering talks to them.
What actually happened was that a bug took out node 1, and so node 2 took over node 1's workload and immediately crashed in the exactly the same way. Unsurprisingly, adding more nodes (beyond 2) didn't make the system any more reliable: they just produced a bigger and less predictable boot loop.
Not every failure would cascade across the cluster like this, but enough bugs did to make the infinite-MTBF claim laughable.
Edit: typo.
@Jplonie @jalefkowit
You're not joking. I used to work for a company that made redundant systems. (It wasn't Tandem.) They were pompously advertised as having infinite MTBF, because marketing people are like that and they plug their ears whenever Engineering talks to them.
What actually happened was that a bug took out node 1, and so node 2 took over node 1's workload and immediately crashed in the exactly the same way. Unsurprisingly, adding more nodes (beyond 2) didn't make the system any more...