[Ksummit-2008-discuss] Suggested Topic: Checkpoint/Restart design

[Eric W. Biederman]

Dave Hansen <dave at linux.vnet.ibm.com> writes:

> On Fri, 2008-08-01 at 15:55 -0500, James Bottomley wrote:
>> How does this relate to the many efforts that went before it (not just
>> on linux, but on a lot of the old unixes plus the posix1003.1m standard
>> that never materialised)?

The big huge gigantic difference is between what previous efforts have
done (especially SGI Irix, Cray and Posix) and what we are doing in
Linux is that we have a solution to one of the most recurring and hard
problems with restart.  We have namespaces so we can guarantee we can
restore global identifiers without collision.  Which means we don't
need to worry about return -EBUSY just because some other process has
one of our pids.

> I don't think it relates much more than we're trying to address the same
> application needs.  I went back and looked at how IRIX went and did this
> a couple of years ago (via some manpages on the web), but I don't think
> it is really influencing how we're going forward at this point.


> Every one of the existing Linux checkpoint/restart projects has had as
> one of its goals to be as non-intrusive in the kernel as possible.  Most
> have existed as loadable modules that implement device drivers and do
> 99% of their work with an ioctl().  We're trying not to relate to those
> too closely. :)

However we do have the developers of several of those previous projects
working on the mailing list and most likely contributing code as well.

So we should be successor in spirit if not in implementation.

> The POSIX proposal (crd.lbl.gov/~jcduell/papers/LBNL-49659.pdf) looks
> mostly reasonable, and this proposal would fit in well with what they've
> described there.  With a quick scan through there, I don't see anything
> that would be a show stopper.

????  That doesn't look like a POSIX proposal at all.  I believe IRIX was
roughly compatible with the posix draft.

Eric