Robust Hyperlinks and Locations

We suggest that building \”permissive, but robust\”
digital library systems and services is an attractive
alternative to the library and computer science tradition
of building \”strict, but fragile\” systems. Strict, but fragile,
systems are efforts to engineer complete systems that
ensure desired properties, but which often prove
impractical in distributed environments without a
central authority to coordinate change. In the
permissive, but robust, approach, we permit individual
components to change in ways that might, in fact,
cause a desired property to fail to persist. However, we
engineer components to be robust, so that it is likely
that desired properties persist even under a great deal
of uncoordinated change.

We have applied the permissive, but robust, approach
to two related problems of reference in distributed
information systems. The first application yields robust
hyperlinks, and the second, robust locations. Robust
hyperlinks address the familiar issue of providing
persistent reference to networked resources, such as
Web pages, given changing, uncooperating services.
Robust locations concern a somewhat less familiar,
but, we suspect, soon-to-be just as big a problem,
namely, references to changing sub-document
resources. Robust locations, we suggest, provide
essential grounding for next-generation web
functionality, such as annotations that survive
document editing.

In both cases, robustness is achieved by providing
multiple, independent descriptions across boundaries
where change is likely to be uncoordinated. If the
different descriptions are property selected, then most
uncoordinated changes will be unlikely to cause all the
descriptions to fail. Thus, while there is no guarantee
that references will remain coherent, a single failure is
unlikely to be catastrophic. Instead, the failure of one,
even a primary, method will generally allow graceful
recovery via other methods.