Leverage the features of IBM GPFS cluster file system to store cached images and instances

IBM General Parallel File System (GPFS) is a mature cluster file system used by some of the largest enterprises and supercomputers in the world. It offers a number of specific features and optimizations for hosting master images, instances, and volumes. In particular, the bock-level format-agnostic copy-on-write mechanism enables quick instance provisioning that avoids data copy. The File Placement Optimization (FPO) feature allows specifying the set of nodes and their local disks where the physical blocks of a particular image file and its replicas should be allocated. Transparent block-level replication, controllable per file, provides resilience.

This BP proposes a small set of changes to leverage the features of GPFS to optimally store cached images and instance files.

GPFS is already supported as a Cinder backend, and GPFS as a glance store blueprint have been approved. GPFS as instance store is meant to provide a better integration with nova when GPFS storage is used