The idea is to use sqlite3 to work around the disk size constraint for individual objects and/or their attributes (key-value pairs) so that the Ceph implementation can manage/stripe the underlying data blocks across OSDs seamlessly. The implementation would be an sqlite3 VFS back-end interface to Ceph. eg. metadata server (MDS) could create an sqlite3 database using this new back-end to manage objects spread across OSDs So, do I use the RadosStriper or the Filer class for the implementation ? Or, do I need to tweak current implementation of some class(es) to get the desired functionality in place to be used in the sqite3 VFS interface implementation. -- Milind
