Delivery-Date: Wed Sep 24 22:43:31 2008
Date: Wed, 24 Sep 2008 22:51:38 -0700
From: Leslie Cheung <lccheung@usc.edu>
Subject: verifying your simulation
To: cs551@merlot.usc.edu
Message-id: <019f01c91ed2$c7acdd10$57069730$@edu>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii
Content-language: en-us
Content-transfer-encoding: 7bit
Thread-index: Acke0scOiczZDxyDQC2FZeFzYfIrNQ==

Hi all,


A few of you have asked me how you can verify if your simulation is
"correct". Here are a few things you should try.

1) Make sure the statistics match the simulation output. If you have 2
customers, and inter-arrival time of c1 and c2 are 1 and 0.5 respectively,
the average inter-arrival time should be 0.75, and not, for example,
0.743123. I recommend calculating the statistics by hand using the
simulation trace, and compare them with what you output at the end.

2) Check the simulation with "real" time (e.g., a clock). During my office
hour, I saw a student's simulation finished 20 customers in about a second
using the default parameters. Since the average inter-arrival time is 2
seconds (the default value for lambda is 0.5/second), you should expect, on
average, a customer arrives every 2 seconds in real time. That means the
simulation should pause a little bit after printing a few lines to stdout.

3) Make sure your simulation clock always goes up. The simulation clock
should *never* roll back. For example,

00001520.374ms: c9 arrives
00001625.541ms: c10 arrives
00001524.263ms: c8 departs, inter-arrival time = ...

The time in the 3rd line (c8 departs) is earlier than the time in the 2nd
line (c10 arrives), and this should never happen. If you see something like
this in your simulation, you better find out why and fix it!


Some students also asked me about how they can verify if there is no
deadlock, or if the mutex are put in the correct places. Well, there is no
easy way to do it, other than inspecting your code very carefully. As for
deadlocks, you might want to try running your simulation with a large number
of customers. Just keep looking at your code and/or run simulations until
you convince yourself you are right!


--Leslie