RBF: A New Storage Structure for Space-Efficient Queries for Multidimensional Metadata in OSS

Object-based storage systems (OSS) have evolved from the traditionally unintelligent management to intelligent and selfmanaged schemes by delegating low-level management activities to storage devices, and providing new functionalities, such as greater scalability and dynamic reconfiguration. An object consists of data, user-accessible attributes and device-managed metadata, which is responsible for mapping data structures, such as files and directories, to blocks on storage devices. Such metadata usually contains multidimensional information for representing the mapping relationship with objects that have both physical and logical attributes, such as access time, data size, request amount, access pattern, and QoS agreement. Thus, substantial storage space has to be used to store multidimensional metadata, along with highly complicated operations of querying multidimensional metadata, due to complex data organization structures. Common query operations, mainly including point query and range query that are independent of the underlying data layout, allow one to search objects that have multidimensional attributes, with the help of multidimensional metadata. • Point query determines whether a given object is a member of a data set. For example, a request of point query for multidimensional metadata may wish to know whether an object with “ID=xyz” and “data size=100GB” attributes is a member of the current storage system. • Range query obtains all objects whose attribute values exist within the ranges of a query request. For example, a request of range query for multidimensional metadata may wish to know all the objects having “data size>150GB” and “access time<20:00” attributes.