On Sat, Nov 04, 2000 at 09:35:54AM -0500, Jean-Louis Martineau wrote:
> On Fri, Nov 03, 2000 at 11:23:10PM -0500, John R. Jackson wrote:
> > >I think that's what's happening, a dump with est_kps larger than the
> > >bandwidth will never be scheduled to dump on holding disk.
> >
> > I agree it looks like that could happen in principle (in fact, I wonder
> > if other such "don't do this now" states should be examined), but in this
> > particular case the estimated bandwidth for the two disks not processed
> > was well below the available.
> >
>
> The schedule give:
> --------
> admin1.cor sda9 lv 2 t 5 s 52939 p 13
> --------
>
> 52939/5 = 10587 that's above 10000, and that's why sda9 didn't get dumped.
In free_kps() we have:
/* XXX - kludge - if we are currently using nothing
** on this interface then lie and say he can
** have as much as he likes.
*/
if (ip->curusage == 0) res = 10000;
else res = ip->maxusage - ip->curusage;
Even if maxusage > 10000, it will never return more than 10000 if curusage==0.
Should we fix it and increased it to 100000?
Or my previous patch is better and we should remove this kludge?
Jean-Louis
--
Jean-Louis Martineau email: [EMAIL PROTECTED]
Departement IRO, Universite de Montreal
C.P. 6128, Succ. CENTRE-VILLE Tel: (514) 343-6111 ext. 3529
Montreal, Canada, H3C 3J7 Fax: (514) 343-5834