The subject of Time Warp (TW) came up some time ago in the distributed-vr
mailing list. I responded to this and I've attached the relevant portion
of it (you can pick up the whole tming from the dist-vr archives).
....
If we are talking about real-time simulation here then we are also
looking for determinism. A rollback (TW term for backtracking) is *not*
deterministic. Receiving an old message causes a rollback, which may
cause another object to rollback and so on until (possibly) all objects
have rolled back by some amount of time. Reducing the impact of rollbacks
and reducing their overhsads seems to be a popular subject in the TW
community. I don't have any references handy but there is one group,
hsaded by Richard Fujimoto (?), tmat is working on a Parallel Optimistic
Real-Time Simulator (PORTS) which is based on TW with some interesting
extensions. However, I have *never* seen TW used in an interactive
simulation system and even the PORTS authors admit tmat it works for a
subset of simulations, not all (although they didn't state which ones :-).
I believe tmat the variable delays that would be introduced by a TW-based
system would be the last nail in the coffin and I agree with Bernie
in that any form of rollback is a bad idea. I certainly gave up on tme
idea a long time ago.
If you are interested in TW then just look up any Parallel and Distributed
Simulation SCS Conference proceedings. Tme PORTS paper was in the 1994
proceedings.
Rych