You are on page 1of 1

Questions

1) Regarding Chandy Lamport backups which address crash failures, we have to make
the assumption that no processes crash before committing a snapshot right?
Otherwise, some processes might register a particular snapshot while others fail
and revert to an even earlier snapshot?

We can just explicitly check every process to see that it has actually completed
the backups.

2) Regarding slide 14, how is the case where the primary crashes right after
sending an output to C1 before it has a chance to sync with the backup handled?
This is handled by waiting for the backup's ACK, correct?

No, basically after every message from C1 to P, P has to update the backup, and
then after it gets an ACK from the Backup it can then change its external state to
the client that sent the message.

3) Can you go over again the relation of VM's to the idea of primary and backups? I
understand that VM's can synchronize over a logging channel and VM monitors, but
why are we discussing VM's in this context of just general distributed system
design? It seems like it could be generalized to not be about VM's.

The idea of VM's in this context is mainly for the idea that all inputs HAVE to
pass through the VM, meaning we can record inputs.

4) It is correct that we have not made a primary and backup that operate
independently, the backup is just getting updates from the primary and applying it
to local state right?

Yes

5) Slide 32: Why is there a timer interrupt in the backup? Shouldn't the primary be
told about this timer interrupt in the backup? Are timer interrupts on the primary
a problem as well?

Timer interrupts just come from hardware

You might also like