Risk Pooling, Late Customization and Product Commonality

This guest post is one of a series on risk pooling and its implications. It was authored by Professor Hiroshi Ochiumi, Assistant Professor of Clinical Data Sciences and Operations at USC. Read more about Professor Ochiumi.

The new Apple Mac Pro packs a great amount of features in a very small space. One reason Apple was able to do it is the way the Mac Pro handles the heat. Computers, particularly the powerful ones such as the Mac Pro, generate a lot of heat. Traditionally, computer makers put a heat sink on the CPU, GPU, and other parts to distribute heat. Then they have to design airflow so that there is enough air to cool the heat-generating parts. A typical PC has multiple heat sinks and fans in it. Apple designed the new Mac Pro with just one heat sink, which they call a thermal core. Instead of multiple heat sinks and fans, the new Mac Pro has one aluminum triangular prism in the center, with the CPU board and two GPU boards on the three sides.

How can Apple make the Mac Pro smaller by having just one piece of heat sink (or thermal core) instead of many?The answer: because it is unlikely that all the three boards are working at their capacity at the same time. Chances are one board is generating heat while other boards are not working hard. So the thermal core doesn’t have to be as large. Therefore Apple can manage to make the new Mac Pro smaller.

Since multiple heat-generating parts are utilizing one heat sink, we might refer to this configuration resource sharing. We can also call it a CPU heat sink that doubles as a heat sink for the GPU.  In Operations Management, this is generally referred to as risk pooling.

In the following series of articles, we present the concept of risk pooling and other related ideas. Specifically, we will go over:

  • Location risk pooling
  • Late customization/postponement
  • Product commonality

to see how companies are applying the idea of risk pooling in different scenarios.

The concept of risk pooling

Simply put, the idea behind risk pooling is:

“If we combine multiple streams of randomness, something good will happen.”

In the above Apple story, they used to have many heat sinks which generate heat rather randomly, in the sense that not all the parts are generating heat at one time. Now they combined those heat sinks and something good has happened – the new Mac Pro is very compact despite the great performance.

To illustrate further, let’s consider the following two scenarios for waiting lines.

Scenario 1: We have only one line for multiple servers. Customers form one line and the customer at the very front gets to see the “next available agent.”  This configuration can be seen at the airport check-in counters as well as in bank offices. It is sometimes called a fork system.

Scenario 2: We have one line per server. Customers get to choose which line to join. Your line may or may not move as fast as you wish. We see this in a typical grocery store.

One interesting question is whether one waiting system works better than the other. For the sake of fairness, let’s assume that the only difference between the above two scenarios is the way we form waiting lines. That is, customer arrival rates are identical, passenger check-in takes the same amount of time as grocery check out, and the number of airport ticket attendants is the same as the number of grocery cashier.  It turns out that scenario 1 (the airport system) works better than scenario 2 (the grocery system). In other words, one would expect to wait less in scenario 1 than in scenario 2. Also, on average, there are fewer people waiting in scenario 1 than in scenario 2.

What is going on? After all, arrival pattern is the same and service time is the same. How can we achieve better results just by forming one long line instead of many short ones?

What seems to be happening here is the following. As we all know from our own experiences, if we have to pick one line at a grocery store, sometimes the line we pick moves very fast and sometimes very slowly. At the airport, you can’t get lucky and pick the fastest moving line, but you can’t get unlucky and pick a slow line, either. Things seem to smooth out in scenario 1, and as a result, we expect to wait, on average, shorter in scenario 1 than in scenario 2.

Then why would grocery stores not switch to the airport system? Some grocery stores are actually switching to the airport system as was reported by New York Times.

If financial portfolio management comes to your mind, you are on the right track. The basic idea underlying financial portfolio management is the same. A stock may go up or down. If we invest in multiple stocks, not just one stock, then the risk will be lower. Here again,

“If we combine multiple streams of randomness, something good will happen.”

Next time we will look at more risk pooling applications.