IIRC, pages read from SPOOL-resident DCSSs are counted as paging I/O, so the
scenario you describe in your third paragraph is certainly a possibility.
Marty
> -----Original Message-----
> From: The IBM z/VM Operating System
> [mailto:[EMAIL PROTECTED] On Behalf Of Bill Holder
> Sent: Monday, May 19, 2008 3:30 PM
> To: [email protected]
> Subject: Re: Mystery paging
>
> The dataspace usage is certainly a good thing to check, but there are at
> least a couple of other possibilities:
>
> The paging rate numbers are across time, whereas the Q ALLOC PAGE output
is
> instantaneous. It could be that there's a low volume of paging to/from
paging
> DASD going on, but the pages that are read in are immediately changed,
which
> would cause CP to discard the copy out on paging DASD when CP notices the
page
> change. So it needn't be the same page read 13 times a second, it could
be a
> bunch of different pages which are then immediately changed, for which CP
> then discards the paging DASD copy.
>
> I'd have to check, but it seems possible that demand paging of spool pages
> from (and even possibly to) spool volumes could account for this, I'd have
to
> check how those get counted. In particular, if you have a very lightly
used
> NSS or DCSS, that only tends to be loaded by one user at a time, and it is
> frequently purged and then reloaded, the initial page fault block reads
from
> spool might show up like this.
>
> - Bill Holder
> z/VM development, IBM
>
