On Oct 31, 8:30pm, Jeffrey Hutzelman wrote:
> > Does anyone know what these kind of accesses might be for? In particular,
> > what would cause this one-access-per-volume-per-day pattern?
>
> Careful here... The dayuse data is only updated when the volume actually
> _is_ used. If a volume is unused for several days, or even several
> months, vos will continue to show the number of accessses it had on
> the last day it _was_ accessed.
I checked the AFS 3.4a source code and confirmed this behavior. The
VAdjustVolumeStatistics is called only when a vnode is accessed. This means if
I got an one-access count on a volume for the past day, the last one-access
might have happened long time ago!
This also means that if an HSM scheme using nearline storage based on the
dayUse count might be deceived to keep inactive volume online forever if the
there were a lot accesses done on the last day before it became inactive, e.g.,
after it was backed up by the user with some program that accessed the vnodes.
This indicates that the significant number of low access counts of volumes is
a real issue to be looked at for keeping the storage cost down.
Shyh-Wei
