Handling multiple versions of IBM Storage Scale data

This topic provides description on handling multiple versions of data in IBM Storage Scale.

The primary copy of a GPFS file system and a hardware replicated copy cannot coexist in the same GPFS cluster. A node can mount either the original copy of the file system or one of its replicas, but not both. This restriction has to do with the current implementation of the NSD-to-LUN mapping mechanism, which scans all locally attached disks, searching for a specific value (the NSD ID) at a particular location on disk. If both the original volume and a hardware replica are visible to a particular node, these disks would appear to GPFS as distinct devices with identical NSD IDs.

For this reason, users are asked to zone their SAN configurations such that at most one replica of any given GPFS disk is visible from any node. That is, the nodes in your production cluster should have access to the disks that make up the actual file system but should not see the disks that hold the replicated copies, whereas the backup server should see the replication targets but not the originals.

Alternatively, you can use the nsddevices user exit located in /var/mmfs/etc/ to explicitly define the subset of the locally visible disks to be accessed during the NSD device scan on the local node.

The following procedure is used to define an nsddevices user exit file to instruct GPFS to use a specific disk diskA1 rather than other copies of this device, which might also be available:

echo "echo diskA1 hdisk" > /var/mmfs/etc/nsddevices chmod 744 /var/mmfs/etc/nsddevices

Refer to the prolog of /usr/lpp/mmfs/samples/nsddevices.samples for detailed instructions on the usage of nsddevices.