Random numbers and virtual-machine forks

Posted Mar 11, 2022 18:43 UTC (Fri) by bof (subscriber, #110741)
Parent article: Random numbers and virtual-machine forks

Hmmm. I understand checkpoint+restore as a migration concept, but my mind's coming up blank with scenarios where a virtual machine would actually be forked into two completely identical copies (that now notice they are to be not so identical in a growing number of ways?)

I mean, duplicate MACs? Duplicate IPs? Huh? What do I miss?

Random numbers and virtual-machine forks

Posted Mar 11, 2022 21:14 UTC (Fri) by josh (subscriber, #17465) [Link]

Fast startup: boot your VM to the point where it's about to invoke userspace, then fork it.

Fast per-client VMs: boot your VM to the point where it's listening for connections, then fork it.

You don't need to worry about IPs or MAC addresses if your VMM is handling routing and giving your VM an internal-use-only IP.

Random numbers and virtual-machine forks

Posted Mar 12, 2022 0:58 UTC (Sat) by iabervon (subscriber, #722) [Link]

Could be good for software upgrades for services that have state. Before you start, fork the VM; after you've done the upgrade and tested the system, you can decide which fork to keep.

There are various applications for checkpoint+restore+restore+..., where the copies are mostly not contemporaneous; it's really nice for being able to produce test environments that aren't full of the remains of previous testing, for example, but you generally don't need to have your pristine copy running all the time and being forked into new copies while live.