[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

"Going backwards" & different schedulers




I posted last month a question asking for general or specific guidelines
for fixing a "scheduler going backwards in time" problem.  The most
useful advice I received was to "try a different scheduler."  At the
time of my posting, I was using the Calendar scheduler.

Since then, I have found that my simulation (which uses some C++ code I
have written) works fine (=does not "go backwards in time") using the
Heap scheduler, but *not* with the List or Calendar schedulers.  I made
my comparisons changing only the scheduler, but kept everything else (in
particular, the random number generation seed) identical.

This indicates to me that, despite the claim of the "NS Notes &
Documentation," the three aforementioned schedulers do not always yield
identical results for identical inputs, at least not for my simulation.

(Some users said the problem with "going backwards" may be magically
alleviated by using fewer and/or slower traffic sources.  This suggests
a problem dealing with a large number of [pending] events.  Indeed, in
my simulation, at any given time I expect there may be thousands of
events pending.  So perhaps the "going backwards in time" problem may,
in some cases, be related to having uncomfortably many pending events
managed by the scheduler.)

I post this information to the mailing list simply in the hope it may
help others.

Good simulating,
--daniel

---------------------------------------------------------------------------
 Daniel Friedman    [email protected]    http://www.isr.umd.edu/~danielf