Re: [9fans] Remus: High Availability via Asynchronous Virtual Machine Replication

Re: [9fans] Remus: High Availability via Asynchronous Virtual Machine Replication

[erik quanstrom]

On Thu Dec  3 15:13:17 EST 2009, roman@shaposhnik.org wrote:
> This appears to be an interesting way of solving an HA problem:
>      http://nss.cs.ubc.ca/remus/papers/remus-nsdi08.pdf
>
> It is also different from a typical approach of checkpointing in HPC.
> One thing that I'm wondering about though is whether Plan9 architecture
> of things like /net and /dev/sd would actually allow it to be HAed in that
> fashion *without* a help of the hypervisor.
>
> Does it make sense or am I totally off base here?

i don't see why it wouldn't work with sd.

networking would be harder.  since the stated goal
is to not reset tcp connections, you'd need to force the
backup to pick the same ports and starting sequence numbers
as the primary.

- erik

[9fans] Remus: High Availability via Asynchronous Virtual Machine Replication

[Roman Shaposhnik]

This appears to be an interesting way of solving an HA problem:
     http://nss.cs.ubc.ca/remus/papers/remus-nsdi08.pdf

It is also different from a typical approach of checkpointing in HPC.
One thing that I'm wondering about though is whether Plan9 architecture
of things like /net and /dev/sd would actually allow it to be HAed in that
fashion *without* a help of the hypervisor.

Does it make sense or am I totally off base here?

Thanks,
Roman.