8

Do dual CPUs on a modern server actually provide redundancy? If one fails can the server be rebooted and run on the remaining CPU?

In the past I have come across the case where CPU 0 always had to be present so if CPU 1 had failed it would reboot fine but if CPU 0 failed you had to physically swap CPU 1 intyo the cpu 0 socket. I would hope this restriction is no-longer present in modern dual CPU servers but wanted to confirm that.

I have actually been told by HP that dual CPUs do not offer any redundancy because they both have to be present but not sure that is correct.

NickC
  • 2,313
  • 13
  • 40
  • 52
  • 2
    A computer/server does not have dual processors for the purpose of redundancy. There's no such thing AFAIK as redundant CPU's. – joeqwerty Jan 16 '14 at 16:17
  • 2
    @joeqwerty http://en.wikipedia.org/wiki/Tandem_Computers – Chris S Jan 16 '14 at 16:27
  • 1
    @joeqwerty Just because you don't know about it doesn't mean it doesn't exist. – MikeyB Jan 16 '14 at 16:33
  • 1
    @MikeyB - That's why I said AFAIK (as far as I know). I didn't say that I knew definitively that they didn't exist. Chris S provided a link to more information and now I know that they do/did exist. – joeqwerty Jan 16 '14 at 16:37
  • For what it's worth, though, I do understand this to not be the usual case or even something most people would encounter. Of course it can be done. – Falcon Momot Feb 15 '14 at 06:47

2 Answers2

8

Do dual CPUs on a modern server actually provide redundancy?

Yes, they might. For example, Processor Instruction Retry is available on Power systems and is capable of recovering from a failed processor by retrying the instruction that failed during execution, possibly on a different processor.

This isn't available on low end servers such as Intel or AMD, however.

If one fails can the server be rebooted and run on the remaining CPU?

Yes, as long as the server hardware is able to properly detect and recover from the failure. I have a lot of experience with the IBM Intel servers and they are pretty good about recovering from a failed processor and allowing the system to boot from the other socket.

MikeyB
  • 38,725
  • 10
  • 102
  • 186
  • 4
    This is one of the many reasons I worship at the altar of IBM Power :-) – voretaq7 Jan 16 '14 at 16:38
  • In this case customer wants a Dell or HP. Dell claim that their servers will reboot and run in the event of one CPU failure, HP on the other hand don't think that their server will. However neither of them really seemed to be certain. – NickC Jan 16 '14 at 17:11
  • They *should*. I have experience that the IBM Intel servers do, but not 100% of the time depending on the failure. – MikeyB Jan 16 '14 at 17:56
2

Do dual CPUs on a modern server actually provide redundancy?

No.

If one fails can the server be rebooted and run on the remaining CPU?

Possible. Depends. If it is dead enough that it is not seen then the Bios may assume only one CPU is in, which may be supported (it mostly is) though yo may loose RAM access (depending on what CPU - some route half the RMA through each CPU).

It alls is depending on how the mOBO handles the situation if a CPU is not there - which is an implemenetation detail.

That said, this is a comically rare case for a CPU to fail in a proper server.

TomTom
  • 50,857
  • 7
  • 52
  • 134
  • 4
    I've seen servers old enough to drink, never seen a CPU go bad. – Chris S Jan 16 '14 at 16:31
  • Are modern CPUs really that reliable now. Over the past twenty years or so I have seen maybe a dozen Intel CPUs die in service, admittedly never seen an AMD failure but in this case the customer wants Intel. – NickC Jan 16 '14 at 16:33
  • @NickC We see approximately a 0.5% processor failure rate per year across a LARGE Intel installation. – MikeyB Jan 16 '14 at 16:36
  • @ChrisS Agreed - in my experience, if the CPU isn't DOA it'll be fine forever. – Dan Jan 16 '14 at 16:39
  • Exactly. That is my experience also. Give it a good burn in and it will not fail. Especially as proper server have passive CPU coolers anyway ;) – TomTom Jan 16 '14 at 16:43
  • Correct answers for Intel servers. – mfinni Jan 16 '14 at 16:55