At Wed, 30 Jun 2004 20:21:01 +0000 (UTC), "Erik E. Fair" wrote:
> It's very important to separate Time of Day from interval timing.

There's one more category of "time" that's probably worth noting.  (I
quote "time" because it may not necessarily be an actual time value.)

        timestamps (or similar) for global ordering

(on multi-cpu systems, it might not be 'global' -- there might also be
a per-cpu value, or a per-node value.)


And example is:

        A given file system operation might want to be tagged with a
        'time' for global ordering (i.e., so that if time changes,
        operation is not considered to have occured before another
        operation when sorting/batching i/o).

        The same file system operation might want to mark a file with
        current Time of Day (e.g., modify time).

I suspect that trying to use an actual time value and forcing it to
always increase was an attempt to shoehorn both of those together.

Obviously, that causes some problems.  8-)


cgd