While reading a paper, also pay attention to the authors and why they encountered this problem.

While replicating, there are 2 options for replication:

  1. Sending the input to the replicas
  2. Sending the changes to the replicas More often than not, sending the input is more performant and easier to do.

VMware hypervisor is a binary translation hypervisor(Another hypervisor type is TRAP)

Incase of this paper, not every single state(of the VM) is replicated, but only some key states are recorded and replayed(as the performance tradeoff is too much).

Thread scheduling is deterministic for single core systems(mostly). For multi-core systems thread scheduling is non-deterministic(because threads can be scheduled to run simultaneously on two different cores).


Search up hypervisor types and levels.