Failures

​ fail-stop failure

not covered: (software / hardware) bugs, design defects.

Whether its worthwhile. –expensive. it’s an economic question

Ways of replica

State transfer

  • Primary sends.

Replicated state machine

  • send external events. arriving Input. Operation from clients. (otherwise, the executions are deterministic function)

  • usually small

  • complicated. rely on sort of assumptions

How to close sync if primary fails?

cut-over, anomalies

new replicas are hard to create.

What state to replicate?

application level replication if more efficient

level of machine? memory & registers.

  • downside: slow

VMware FT

Non-deterministic-events

​ Inputs-packet-date + interrupt

​ Weird instructions

​ Multicore