IBM General Parallel File System (GPFS) is a cluster file system that provides concurrent access to file systems from multiple nodes. The storage provided by these nodes can be direct attached, network attached, SAN attached, or a combination of these methods. GPFS provides many features beyond common data access, including data replication, policy based storage management, and space efficient file snapshot and clone operations.

The GPFS driver enables the use of GPFS in a fashion similar to that of the NFS driver. With the GPFS driver, instances do not actually access a storage device at the block level. Instead, volume backing files are created in a GPFS file system and mapped to instances, which emulate a block device. GPFS software must be installed and running on nodes where Block Storage and Compute services run in the OpenStack environment. A GPFS file system must also be created and mounted on these nodes before starting the cinder-volume service. The details of these GPFS specific steps are covered in GPFS: Concepts, Planning, and Installation Guide and GPFS: Administration and Programming Reference. Optionally, the Image Service can be configured to store images on a GPFS file system. When a Block Storage volume is created from an image, if both image data and volume data reside in the same GPFS file system, the data from image file is moved efficiently to the volume file using copy-on-write optimization strategy.

To use the Block Storage service with the GPFS driver, first set the volume_driver in cinder.conf: volume_driver = cinder.volume.drivers.ibm.gpfs.GPFSDriver The following table contains the configuration options supported by the GPFS driver. The gpfs_images_share_mode flag is only valid if the Image Service is configured to use GPFS with the gpfs_images_dir flag. When the value of this flag is copy_on_write, the paths specified by the gpfs_mount_point_base and gpfs_images_dir flags must both reside in the same GPFS file system and in the same GPFS file set.

It is possible to specify additional volume configuration options on a per-volume basis by specifying volume metadata. The volume is created using the specified options. Changing the metadata after the volume is created has no effect. The following table lists the volume creation options supported by the GPFS volume driver.

Volume Create Options for GPFS Volume Drive

Metadata Item Name	Description
fstype	Specifies whether to create a file system or a swap area on the new volume. If fstype=swap is specified, the mkswap command is used to create a swap area. Otherwise the mkfs command is passed the specified file system type, for example ext3, ext4 or ntfs.
fslabel	Sets the file system label for the file system specified by fstype option. This value is only used if fstype is specified.
data_pool_name	Specifies the GPFS storage pool to which the volume is to be assigned. Note: The GPFS storage pool must already have been created.
replicas	Specifies how many copies of the volume file to create. Valid values are 1, 2, and, for GPFS V3.5.0.7 and later, 3. This value cannot be greater than the value of the MaxDataReplicas attribute of the file system.
dio	Enables or disables the Direct I/O caching policy for the volume file. Valid values are yes and no.
write_affinity_depth	Specifies the allocation policy to be used for the volume file. Note: This option only works if allow-write-affinity is set for the GPFS data pool.
block_group_factor	Specifies how many blocks are laid out sequentially in the volume file to behave as a single large block. Note: This option only works if allow-write-affinity is set for the GPFS data pool.
write_affinity_failure_group	Specifies the range of nodes (in GPFS shared nothing architecture) where replicas of blocks in the volume file are to be written. See GPFS: Administration and Programming Reference for more details on this option.

This example shows the creation of a 50GB volume with an ext4 file system labeled newfs and direct IO enabled: $ cinder create --metadata fstype=ext4 fslabel=newfs dio=yes --display-name volume_1 50

Volume snapshots are implemented using the GPFS file clone feature. Whenever a new snapshot is created, the snapshot file is efficiently created as a read-only clone parent of the volume, and the volume file uses copy-on-write optimization strategy to minimize data movement. Similarly when a new volume is created from a snapshot or from an existing volume, the same approach is taken. The same approach is also used when a new volume is created from an Image Service image, if the source image is in raw format, and gpfs_images_share_mode is set to copy_on_write.