Rodger and Paul,
Do you think the declarative or time-implicit nature of ActiveVRML's
design makes the problem of synchronization harder? In other words,
would an imperative or more time-explicit approach diminish these problems?
- Conal
----------
| From: Rodger Lea <rodger@csl.sony.co.jp>
| Date: Saturday, December 09, 1995 12:41AM
|
| [deleted text]
|
| If I understand your model correctly then the problem Paul is trying to get
| at (I think) is that your approach of hiding inconsistencies during a given
| time interval and then going back and repsiring those inconsistencies will
| only work in a closed system.
|
| Your model seems to propose some notion of rollback, rollback only works in
| a closed system where the effects of the incorrect assumptions (the
| inconsistencies) are contained or bounded, Usually this is done in within
| the scope of a transaction.
|
| A shared VRML world is not a closed system, it exposes its internal state
| to the user and to the rest of the world users via the network. As such, it
| becomes very difficult to rollback - in your text I think you suggest that
| its possible to rollback a local world view without the user becoming too
| confused. I don't agree with that and I think you are making an implicit
| assumption that the time periods during which inconsistencies occour will
| be short. In a shared 3d world with users spesad out across the network you
| will end up with long periods of inconsistency, and with different
| inconsistencies - distributed rollback is a difficult problem.
|
| If I understood Pauls original comment - the issue is that your functional
| model hides the notion of time (or abstracts away from it). This makes it
| very difficult to deal with distributed models were each end point has a
| different view of that time.
|
| rodger
|
|