ADL is employing distributed hierarchical storage to store its own collections. The storage and management of the data is simplified by the fact that it is predominantly ``write-once''. Initial storage facilities are small, comprising 150 GBytes of disk storage, both magnetic and optical, at UCSB and approximately 500 GBytes of robotic tape storage at San Diego Supercomputer Center. Other storage sites are currently being added, while access is being provided to collections that are not owned by ADL.
The issue of delivering large items to users is partly resolved with the use of a wavelet-based progressive delivery of items (see below). In cases in which a user does not require information at all levels of resolution in an image, the user need only transmit information down to the levels of resolution required.
We are employing an initial resolution of the issue
of oid's for DOBJ's, based on suggestions of [7],
in a test of interoperability
between ADL and the UC Berkeley DL
.
Since a goal of the interoperability test is
to operate without knowing
the details of each other's storage model,
storage interoperability involves the implementation of a server
that accepts the handles and returns objects.
In this model, a DOBJ is a data structure
whose main components are data and key metadata.
The essential component of
the key metadata is a globally unique identifier known as a handle.
Handles are provided by authorized handle generators
and, if the handle of a DOBJ is made known to a system of handle
servers, the registered DOBJ may be stored in
a repository. Repositories are also named by a global naming authority
and possess mechanisms for adding new DOBJ's to
their collections and for making the objects available
with the use of a repository access protocol.
Hence, to retrieve a registered DOBJ a user must, at the very least,
present a handle to a handle server to learn the network names
or addresses of repositories in which the corresponding DOBJ
is stored.