[AFS3-std] Re: first draft: ubik update proposal

[Andrew Deason]

On Mon, 14 Feb 2011 17:23:58 -0500
Jeffrey Hutzelman <jhutz@cmu.edu> wrote:

> > struct ubik_nversion {
> >     afs_int64 epoch;
> >     afs_int64 counter;
> > };
> >
> > struct ubik_ntid {
> >     afs_int64 epoch;
> >     afs_int64 counter;
> > };
> 
> 64-bit epochs are necessary.
> 64-bit counters seem excessive.

We don't "need" them, but the extra 4 bytes don't seem like a big
problem. We know 32-bit counters do rollover at some sites. Although
that shouldn't be a problem (once that one related bug was fixed in
OpenAFS), an epoch/counter combination should ideally represent a unique
version of the database or a unique transaction, shouldn't it?

-- 
Andrew Deason
adeason@sinenomine.net