[MUD-Dev] Atomic functions

[Felix A. Croes]

J C Lawrence <claw at kanga.nu> wrote:

>[...]
> I do, or did, pro-actively check, by maintaining interest-party
> lists for all objects in-cache which are currently being processed
> by a in-progress transaction (which I called an "event" for silly
> historical I-didn't-know-any-better reasons).  Then, upon commit I
> pinged all the transactions which had announced interest in the
> objects to which I committed changes.  It was then up to the
> transaction in question to fall on its sword or, umm, pray
> forlornly.
>
> No rocket science, pretty simple code in fact, nut kinda cute.
>
> Was it worth it?  Probably not.  There are a number of obvious
> optimisations that could be made however to reduce the overhead on
> things like the number of affected objects, timestamp of last commit
> to the object etc, but I haven't looked into these or attempted to
> graph any of the models.

James Anderson is doing some interesting research on the subject:

    http://www.cs.unc.edu/~anderson/papers.html

and especially

    http://www.cs.unc.edu/~anderson/papers/ieeepds99.ps

Regards,
Felix Croes

P.S. This doesn't have much to do with atomic functions, though :)

_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev