I am interested in some technical clues on how a failure on a RAM that is shared between multiple computation units (this could be virtual machines managed by hypervisor or two (or more) OSes running on two (or more) physically different computation units, that share the same RAM) can affect the behavior of the functions that run on these different computation units.

More precisely, what are the possible errors that can be provoked by a RAM failure when the RAM is shared between multiple systems? Can this failure be undected?

  • do you have a smp setup? I think that can be detected through that. Maybe look into scalemp? We need more information on how the cluster is setup. – Tux_DEV_NULL Apr 13 '18 at 9:49

Short answer: anything.

Whether the memory is shared or not, memory corruption trashes your data (or even the code) and renders the result completely useless. When the code is hit the application is likely to go haywire and might even trash more data. At the very least it'll crash more or less gracefully.

To avoid this, you might want to look into ECC memory.

  • Thanks for your answer. Actually, I am looking to find different kind of failures that could be provoked only by RAM failures, in order to abstract them in a model. TBH, I also think that this may be too broad to be clearly identified. – Ecterion Apr 13 '18 at 19:44
  • What kind of failures? Today, not only simple Hamming code is used but also more sophisticated techniques including chipkill and memory RAID, dealing with the loss of a chip or an entire DIMM. – Zac67 Apr 13 '18 at 19:51
  • Well, that's my question actually. I don't know what kind of failures are common in a RAM. Bit flip? Wrong access? – Ecterion Apr 17 '18 at 14:34
  • Single bit flips, row or column faults (rare), addressing faults, single chip death, total DIMM failure - all addressed I think. – Zac67 Apr 17 '18 at 17:18
  • Thanks, that is what I was looking for. Do you have any resources related to how these failures could appear (the conditions) and how are they identifiable? – Ecterion Apr 18 '18 at 14:39

Your Answer

By clicking "Post Your Answer", you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.