🚀 go-pugleaf

In article <1992Sep4.131659.122@inet.d48.lilly.com>, rz63633@stewball.d50.lilly.com (Gregg A. Mantock) writes:
> Does anyone know if a file can span accross multiple devices
> in a clustered file system?  Say that a file is created on a
> particular device in a clustered file system.  As the file
> continues to grow it fills the device that it was created on.
> Will the remainder of the file be written to one of the other
> devices in the cluster, or will the process writing to the 
> file abend?
> 
> This becomes an interesting situation when the Data Migration
> Facility and File System Monitoring are running on the
> clustered file system.  Fsmon shows %used based on the whole
> cluster of devices, but df shows the %free on each device.
> Can I assume that fsmon knows what it is doing and that the
> clustered file system need not be migrated until the entire
> cluster hits the warning/critical threshold?
> 
> Thanks,
> Gregg
> -- 

Yes, the clustered file system need not be migrated until the entire
cluster's warning/critical thresholds are reached.  DMF manages a clustered
file system as it would a normal (i.e. contained on one device) filesystem.
Therefore, you want DMF actions started by fsmon, since it also views
the cluster as a single filesystem. 

Hope this helps,
Jim
-- 
--------------------------------------------------------------------------
Jim Gmeinder		Cray Research, Inc.		(612)-683-5581

Gregg A. Mantock wrote...

> Does anyone know if a file can span accross multiple devices
> in a clustered file system?

[text deleted]

> This becomes an interesting situation when the Data Migration
> Facility and File System Monitoring are running on the
> clustered file system.

...and in response, Jim Gmeinder wrote...

> Yes, the clustered file system need not be migrated until the entire
> cluster's warning/critical thresholds are reached.  DMF manages a clustered
> file system as it would a normal (i.e. contained on one device) filesystem.
> Therefore, you want DMF actions started by fsmon, since it also views
> the cluster as a single filesystem. 

...which ducks the question a bit.  If a file could not span
devices, it could theoretically fill its partition without
filling the file system past the fsmon trigger point.  Migration
wouldn't be triggered although from the user's point of view
there would be no more disk space (at least for that file).

But on a Cray 2 (NC2FS), which I assume Gregg is working with
because he refers to clusters, a file can span devices.  If the
original partition fills up space will be allocated from others
in the cluster.

I've had problems where a clustered physical device fills up and
causes problems, even though the file system has free space on other
partitions.  But only when restoring dumps.

David McNab
mcnab@nas.nasa.gov
Parallel Systems Support
NAS Program, NASA/Ames Res. Ctr.