Re: Hurdish applications for persistence

[Jonathan S. Shapiro]

I have just spoken to Marcus to ask for an explanation of what "passive
translators" are. There is some humor here.

As I have said, we considered *not* making Coyotos persistent. We have a
design that is very similar to passive translators that would mostly
work. We have not built it, but I *think* that it resolves most of the
concerns about security that Marcus has identified.

We decided not to do it for two reasons:

1. It would be *much* slower than just keeping persistence.

2. There is a problem of consistency across applications. In a component
system where processes act as components, it isn't good enough in the
general case to restart a process. What you really need to be able to do
is to re-establish a connected graph of processes in such a way that all
of them agree on their current state and their state relationships to
their peers.

For example: an IPC that was in progress at the time of checkpoint must
be restored in such a way that it is still in progress at exactly the
same point in the transfer.


shap