On Mon, 2002-08-26 at 13:06, Andi Kleen wrote:
> On Mon, Aug 26, 2002 at 12:12:16PM -0500, Steve Lord wrote:
> >
> > On Mon, 2002-08-26 at 10:52, Chris Tooley wrote:
> > >
> > > We're running a point of sale system on Linux and have put the data on
> > > an XFS partition. Unfortunately there has been consistent corruption of
> > > the data during writes to the database. Since the database is an ISAM
> > > database, it's all stored in files that are opened and closed a lot.
> > > The vendor of the point of sale is blaming XFS for the corruption. The
> > > application is an old COBOL app that is using the binary compatability
> > > modules to run. Are there any known problems with corruption when using
> > > these binary compatability libraries with 1.1?
> >
> > I am not aware of any specific corruption issues in 1.1, I know
> > nothing of the 'binary compatibility modules you are referring
> > to. Is this for running things like a.out binaries?
>
> Suspect he refers to the iBCS/Linux ABI patches for SCO etc. emulation.
> Christoph should know more about it, he used to maintain that stuff.
>
> I doubt it can cause specific data corruption thought - read/write
> should be directly handed through to the linux implementations because
> there is not much to emulate there.
>
> >
> > Do you have any information about the pattern of the corruption, or
> > the types of I/O being done to the files?
>
> Point-Of-Sale sounds like it'll be often power cycled without proper
> shutdown. This could cause problems if the HD does write buffer a lot of data
> on its own. I would try applying the ordered writing patches or at least
> turn off the write cache of the HD.
>
The application is a linux application and runs in text mode only.
There is no gui. I can cause the same problems when I only allow ssh
access to it as well. That means no reseting because I don't restart
the server.
> -Andi
>
|