STEM help / Training / Two-hour tutorial demo

Exercise 12: Simulating variability between sites

Watch the video presentation and/or read the full text below

Now we are going to think more carefully about how the numbers of customers are spread out between the various locations where equipment is installed:

  1. Double-click the light-blue, location link (or select Deployment from the icon menu for the Access card resource). The Deployment dialog is displayed, as shown below, with the input Distribution = One for One by default. This is what determines the minimal results we have seen so far. The only impact is that, literally, at least one unit will be installed for each site (as soon as there is a non-zero demand). We will explore two of the available alternatives as follows.
  2. Click the Distribution field, and select Homogeneous from the drop-down in the formula bar, as illustrated below. This setting indicates that the customer demand should be assumed to be split evenly between all of the sites. When the first card is full at every site, then another must be installed at every site, and so on.

Figure 27: Changing the Distribution input in the Deployment dialog for the Access card resource

  1. Re-run the model (still skipping warnings) and review the updated results.

Figure 28 Capacities and Utilisation Ratio graphs with Distribution = Homogeneous

As soon as the demand exceeds the initial 400 ports, another 25 units are installed. Such a distribution is unlikely in real life, but is useful as an extreme example.

In practice, it is likely that there will be ‘hot’ sites where the initial capacity is exceeded sooner than Y3, and other ‘cooler’ sites where this happens later. At any given site, as demand increases, the number of slack ports will vary in the range 0–15 inclusive. Given the likely variation in timing across sites, a fair estimate of the slack capacity required at any point in time will amount to ‘half a unit’ per site; i.e., 200 ports. This would have the required installation tracking some way above the pink line for Used Capacity, cutting a more gradual path compared to the simplistic red steps shown above.

  1. Switch back to the Editor, and set the input Distribution = Extended Monte Carlo.
  2. Re-run the model (still skipping warnings) and review the impact on the results.

Figure 29: Capacities and Utilisation Ratio graphs with Distribution = Extended Monte Carlo

As you can, this setting maintains a roughly constant overhead of just over 200 slack ports beyond the number of ports actually in use at any given point in time. Beyond the initial constraint of one unit per site, the used capacity is proportional to the demand, whereas the slack capacity is more or less proportional to the number of sites. This is the most prudent approach; so, if in doubt, use Extended Monte Carlo.

(Beyond the scope of this tutorial.)

By all means experiment with the other two options for the Distribution input:

  • Monte Carlo: as above, but without the initial constraint of one unit per site; suitable if a ‘just-in-time’ deployment model is feasible for the first customer at each site
  • Smoothed Homogeneous: like Homogeneous, but also without the initial constraint of one unit per site, but only really included as an academic example for completeness.

Things that you should have seen and understood

Deployment, Distribution, One for One, Homegeneous, Extended Monte Carlo

 

© Implied Logic Limited