0

For example, I read that while VirtualBox never overcommits memory, KVM has options to allow memory overcommit.

So I wonder if popular cloud providers (GCE, AWS, Azure ...) ever overcommit the VM instance memory (and so guest memory has chance to be swapped out on the host), or do they mlock it? Any info on that?

Edit: I meant KVM as an example where the capability is present. More generally I mean if memory overcommit is possible, by whatever means.

Also, in case overcommit is possible, two follow-up questions, for specific providers:

  • What is the specific behavior on OOM event? For example instance crash or relocation? Or can guest memory be swapped out by host?
  • How to prevent memory overcommit as much as possible? For example, using maxed out instances to own full node, etc?
3
  • Azure doesn’t use KVM
    – Ramhound
    Commented 2 days ago
  • @Ramhound - Thank you. I meant KVM as an example where the capability is present. More generally I mean if memory overcommit is possible, by whatever means. Will clarify post.
    – ron
    Commented 2 days ago
  • In general, when building a virtualization cluster, the recommendation is that RAM is not to be over assigned, but CPU can (and should) be. In larger scale setups I would guess that a significant enough number of machines never actually use the RAM assigned, and so this is a place where values can be overlapped for more efficient resource usage. However, the question you have asked is clearly off topic on this site: We do not know the internal configurations of major/popular cloud providers, and the best way to get the information you want is to ask them yourself. Commented 2 days ago

0

You must log in to answer this question.

Browse other questions tagged .