Subject: Re: NTP loses sync if st driver pushed hard?
To: Thor Lancelot Simon <tls@rek.tjls.com>
From: Matthew Jacob <mjacob@feral.com>
List: tech-kern
Date: 09/16/2001 17:22:14
Do you, perchance, for some really strange reason, use the *block* interface
to the tape drive? I have to admit that my experiences with things wedging up
come from some inadvertant testing with the block version (repeat inf: blush).

I rather suspect that any soft error which causes a check condition will
reduce streaming. We may need to rethink whether or not all errors should be
done by the completion thread.

-matt


On Fri, 7 Sep 2001, Thor Lancelot Simon wrote:

> On Fri, Sep 07, 2001 at 05:59:59PM +0200, Manuel Bouyer wrote:
> > On Sun, Aug 26, 2001 at 09:13:39PM -0400, Thor Lancelot Simon wrote:
> > >
> > > On several different machines, all with "options NTP" and ntp4, with
> > > several different SCSI host adapters, both in 1.5 and in current with
> > > thorpej-scsipi, I've noticed that if I push the 'st' driver hard, two
> > > things happen:
> > >
> > > 1) Output periodically pauses; the result is that the tape doesn't stream
> > >    for more than perhaps 30 seconds at a time.  Whether I have four striped
> > >    SCSI disks feeding a DLT8000 on a separate SCSI bus or a single IDE disk
> > >    feeding a DLT2000XT, I can sustain well above the drive's continuous
> > >    write rate for some time, but inevitably there's a pause that causes
> > >    the tape to stop streaming.
> > >
> > > 2) ntpd periodically must adjust the clock by about 0.5-1.5 *seconds* while
> > >    the tape drive is in use.
> >
> > Do you see complete wedges of the machine (keyboard input stops, etc ..) ?
>
> Only occasionally, and always during a tape error condition; Matt may have
> actually more-or-less explained that problem, too.
>
> Normally, if I do something like write a few GB of data to tape, everything
> proceeds as normal during the backup, but once it's over NTP adjusts the
> clock by 0.5-1.5 seconds at once.
>
> Actually, it may occur during long backups, too; I can't remember.  I'll do
> some more testing tonight.  If it happens _only_ at the end, that might
> implicate the close routine (rewinding, perhaps) though I can't see how.
>
>