Workshop on Middleware

Session Summary

Naming and Addressing

Speakers:

Scribe: Lewis & Bienvenido

The session on Naming started with a 'Quiz'; a set of questions about naming. The request was that everyone should answer 'yes', 'no', or `it depends', and, if `it depends' then to answer "what does it depend on, and how?"

The session was supposed to start with some assumptions:

1. It is necessary to have objects that refer to other objects, services, locations, individuals in a way that allows binding (of information-about and access-to) by name alone.
2. Binding must be fast.
3. Name assignment and binding services are a fundamental part of 'middleware'.
4. Fewer naming and binding systems are better than many, as long as they meet requirements.

• A. Scope
  1. Can or should naming of digital objects be separated from naming individuals, roles, objects, systems, services?
  2. Should naming of static (immutable) objects and dynamic objects be handled separately?
  3. Is it possible or advisable to separate name lookup from discovery/search? Where's the line?
  4. Is the design of the name lookup system separable from the design of caching, replication, mirroring, preservation, timestamp, watermark?
• B. Information in names
  1. Should relationships (version/revision, part/whole) be evident from names? Should version relationships be captured in the names?
  2. Is there a separate name for an object and meta-data-about-object? Is there a relationship between those names?
  3. Is it possible/advisable to separate naming for systems & late binding from naming for human transcription, printing in newspapers, business cards? Internationalization issues?
  4. Why not merge other information into names? E.g., rating, cookie, authentication?
• C. Security issues
  1. Should names be self-authenticating? Error-detecting? Error-correcting?
  2. Who is allowed to issue a name? Are there any attacks by nasty name-givers?
  3. Who is allowed to update the information associated with a name? Now? In 5 years? In 100 years?
• D. Growth and Scale
  1. Is it possible to support autoconfiguration and introduction of new naming and resolution services?
  2. While the network is growing exponentially, must the disk, bandwidth and number of name services all also grow exponentially? Which don't?
  3. Can the same system accomodate objects with temporary and objects with permanent names?
• E. Compatibility
  1. Must naming systems and name lookup include every other naming system, and supply mutual encapsulation?
  2. What legacy naming systems shouldn't be included?

The discussion was wide-ranging, but the notes on this session are sketchy. If you have additional comments on these issues, please send them to masinter@parc.xerox.com.