der Mouse wrote:
> On a slightly different note, has anyone - on NetBSD or otherwise -
> ever tried restoring a kernel coredump?  That is, reading it back into
> core and resuming execution?  If done right, this could give us a
> mostly-MI pause/restore facility, of a sort.  (Yes, there are endless
> potential problems with it, maybe severe enough to make it not worth
> doing.  But I don't know whether they're bad enough in practice to
> render it useless, whence comes the question....)

A past employer did something like this as a means of achieving
fault tolerance. The essence (assisted by hardware for performance
reasons) was to restart a memory image of a live operating system
on another identically configured system. There are certain changes
that would need to be made to the driver architecture (to include
an entry point for recovery/restart, and possibly another for
checkpoint/snapshot). You would also need to reserve a memory region
for the code and data needed in order to manage the reloading and
restarting. Beyond that, it is largely a matter of programming.

While it is in fact possible to restore kernel dumps, is is likely
to be a poor return on investment as a development project. In
practice the restarting maching would need to be identical hardware
to that of the crashed machine (a limit to the usefulness). I would
also not be surprised if there are patents issued or pending from
one or more parties that cover key elements of such mechanisms.

/sjd