Nomenclature: Flexible Name Resolution on a Flat Object Namespace
The confluence of two events, the addition of non-volatile memory (NVM) to the memory bus and the continued exponential growth of data gives way to the need for new storage organization principles. File systems do not lend themselves to the organization and retrieval of large data sets, especially in the scientific world. NVM will further add to the complexity of data organization, creating a memory hierarchy where data will be persistent and addressable directly from the CPU, making file system latencies unacceptable for NVM access. We believe to expose the benefits of NVM, operating systems must not hide persistent data behind a file system, and instead present data as a flat object store. To provide naming and organization to this data, we propose Nomenclature, a naming service to allow for multiple organization methods to coexist. This service facilitates naming techniques that have not been seen before, such as recursive naming and name specific interpretation methods, supported by method agnostic reference count. With this, users of Nomenclature can organize the same data without a unified organization scheme without interference. By separating organization and data placement we are able to allow for lightweight organization in userspace while providing the tools necessary to access complex datasets.
Monday, May 7, 2018 at 12:30 PM
Last modified 24 May 2019