A few very specific types of riddles:

  • If you fold a normal piece of paper 50 times, how thick would it be? [Answer] If the paper is 0.1 millimeters in thickness, it would produce a wad of height 1.13×10^(11) meters, or a little more than 80% of the distance between the Earth and the Sun.
  • Would you rather receive $1M for one month of work or a penny on your first day and each day double it for one month. [Answer] The answer is the penny – doubling every day yields just over $10M in a 31-day month.
  • What did Einstein say was the most powerful force in the universe? [Answer] Compounding interest.

The point of all of these puzzles is that exponential processes are difficult for people to grasp – we all have a tendency to underestimate them.

What does this have to do with backup, deduplication and retention? I find that people either drastically overestimate the savings from deduplication or drastically underestimate those savings.

People who overestimate the savings of deduplication do so because they have too much protected data for their raw storage device. This is exacerbated when you have policies that don’t provide compression before deduplication.

When people underestimate the savings of deduplication, they do so because of the same type of problems human beings have when thinking about exponential processes. Deduplication produces some pretty amazing results if the size of the data being protected is small enough compared to the overall physical storage device. Makes sense, right? The more retention you start out with, the more the chances of common data being discovered. The more common data discovered, the more retention you achieve.

What’s the catch? The catch is that many people get disappointed with deduplication because they try to achieve very high retention rates with relatively small amounts of physical storage. This is one reason that looking at the price per raw terabyte and also looking at compression BEFORE deduplication is so important – because the more space you have, the more common data can be found and thus the more deduplication can help.

Would love to hear from you if you think I’m missing something here or if you have other ideas on how to illustrate these concepts.