Integrating content-based access mechanisms with hierarchical file systems

Burra Gopal, Microsoft (formerly U. Arizona); Udi Manber, Yahoo! (formerly U. Arizona)
Proc. OSDI 99

Idea: Combine traditional HFS access mechanisms (symlinks, directories...) with semantic filesystems.  A SFS provides content-addressability for files using some query machinery, and instantiates semantic directories that contain symlinks to files satisfying query.  In HAC (Hierarchy And Content), users can freely combine both mechanisms.  A user-perceptible invariant is that manual modifications to query results by user (whether adds or deletes) are never modified by HAC without explicit instructions.  The content-based access (CBA) mechanism is selectable; authors used Glimpse, via generic CBA interface, and implemented filesystem as a user-level filesystem.  HAC is substantially (46%) slower than Unix FS but requires little (~5%) extra overhead to store (query) metadata.  Support provided for remote-mounting semantic filesystems, etc.

Problem 1: Scope consistency.  Example: delete files from a parent directory, such that the files in the child directory no longer satisfy some constraint of the query that created the directory structure.  These are dealt with in real time as they occur: queries are re-executed with respect to the new scope.

Problem 2: Data consistency.  Example: the contents of a file change, such that it no longer satisfies some query and therefore doesn't belong in a particular semantic directory.  These are resolved by running an ssync operation at user-specified intervals. 

Relevance: related to navigating informational spaces.  An example of an IIA?