On Jan 19, 2007, at 7:35 AM, Gulfie wrote:
1) The map isn't the terrain, and a model isn't a system. If there
are any errors in your model, then the solution is invalidated.
sure enough. but i guess i am relying on the tools doing the changes to
be defensive. see answer to 2.
besides, i am an optimist; every failure caused by a model error si an
opportunity to improve the model.
2) How does the system deal with failure. If two of the 4 nodes went
down, would the other two degrade gracefully or spaz out? How about
communication failure between the hosts?
that is a different conversation (about designing a fault tolerant
system etc). to the extent it applies here,
my normal attitude is if you encounter an serious error, like a node is
down, just stop,
and run the algorithm again to get another solution.
3) Are you looking for a soltuion, a good solution or an optimal
solution? In the above case any of them would work, but with more
complicated constructs the difference may become important.
don't care. any solution is fine by me, but you are right, different
metrics may become more important later.
luckily, (really) modern planning models can have quite sophisticated
metrics to optimise against.
4) Somewhat relating to point 2. With a more complicated rule based
system it may be harder to tell how close the system is to nolonger
being able to meet it's goals. Or stated a differnet way, the cost
implications of a given change may not be redily apparent or
calculatable.
yup. not much to do about that, except figure out how to measure
"fragility" quantitatively, and incorporate it into
your solution generation. even if the planning tool can't do it
directly, you can always use monte carlo and
generate 100 solutions and externally choose the least fragile.
-gulfie
thanks most kindly
andrew
----
Andrew Hume (best -> Telework) +1 732-886-1886
[EMAIL PROTECTED] (Work) +1 973-360-8651
AT&T Labs - Research; member of USENIX and LOPSA
_______________________________________________
lssconf-discuss mailing list
lssconf-discuss@inf.ed.ac.uk
http://lists.inf.ed.ac.uk/mailman/listinfo/lssconf-discuss
