[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
integer times
From: |
Doug Donalson |
Subject: |
integer times |
Date: |
Mon, 25 Mar 96 13:47:58 PST |
Nelson, Roger, and Chris,
I am starting to run into problems associated with the restriction that time
be
a 32 bit integer. To get around the problem I chose a fractional part of my
time
to be four decimal places and then left shifted my floating point time by the
precision and truncated it. The 4 decimal places is not enough resolution for
research purposes but will do for model validation. Also, without wrap-around,
by
actual time ends at 800,000, again too short for research simulations. I ran
into
two problems as a result of this. At system time 4.25 billion (my time
425,000),
the BLT graph crashed with a message of x axis overflow. I assume that this is
because the X axis (time) only handles signed ints. Sometime thereafter, I got
the following message:
Schedule did not complete soon enough for its scheduled repeat. Schedule.m line
552.
The requsit core dump immediatly followed. I am assuming that this was the
point
where I hit the unsigned int limit. Am I correct that both errors are from
time
overflows and if so do you have any suggestions?
Also, at least from my point of view, SWARM really needs some way to allow
floating point times for those of us who do asynchronous event simulations.
Roger
and I talked about one possibility. This was to have a primary schedule based
on
an integer and a sub-schedule that would be indexed by the integer value and
contain the fractional part of the schedule. Is this in the future of SWARM?
Are any of the other beta sites doing asynchronous event simulations and if so
how
did you handle floating point times?
Cheers,
Doug Donalson
PS I was curious as to how much overhead the SWARM scheduler added to my
simulation. I made an extreamly simple scheduler and ran them against each
other
in my model. First one to 2,000,000 events wins. You'll be happy to know that
the race was a draw!
- integer times,
Doug Donalson <=