Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

The Fedora architecture presents some unique problems and opportunities in supporting hierarchical storage. Even though storage cost is the driving reason for using hierarchical storage, it can provide a solution for other system requirements. We hope in this forum can list those requirements to inform the design for a hierarchical storage integration which fits well as part of the overall Fedora architecture. In addition to the system requirements listed above, the design may need to consider large datastreams, partial reads, updates/versioning, and other features already listed in the JIRA tracker.

Special Aspects of the Fedora Commons Architecture

The Fedora Commons Architecture is most strongly represented by the Fedora Repository. The Repository acts as a spanning (or mediation) layer which encapsulates the way the content is accessed. The architecture is not dependent on the common notion of a "directory of files" which dominates thought on how content is managed including the defacto Web architecture. While applications can still use the Fedora Repository as if it were based on a "directory of files" notion, access to content is virtual and uses dissemination services as access endpoints. There is no guarantee of a one-to-one relationship between a file and a dissemination; dissemination services may be quite complicated. Normally applications should not depend on trying to circumvent Fedora to directly access the file system.

Existing HSMs are largely "file-oriented" and are firmly rooted in the "directory of files" convention though many provide a virtual view of physical storage. Underneath the hood Fedora uses the "directory of files" convention for physical storage of managed content and metadata. In fact, a Fedora Repository can be entirely rebuilt from it managed files. It appears feasible for Fedora to delegate physical storage to an HSM. However, it is likely that the Fedora Repository will require modification in order to provide a robust integration with HSMs since applications really should have no knowledge on how the disseminations are accomplished.

Children Display