Economic Model for Long-Term Storage (EMLTS)

Property Description
ID 10
Name Economic Model of Long-term Storage (EMLTS)
Creator and Funding Developed by David Rosenthal
Status The latest blog post on the subject is from 2011
Purpose To predict and compare the cost of long-term storage over time
Information assets Any kind of digital asset, focus on long-term storage only, binary volume
Activities Archival Storage, Administration
Resources Total cost, no specification of capital cost or labour cost, based on experience from storage providers.
Time Future—up to 100 years
Variables Uses four components: Yield Curves, Loans, Assets and Technologies with variables for purchase cost, running cost, migration cost, service time, further detailed into interest rates, decrease in storage cost per storage unit, a “short-term-ism" factor, planning horizon
Type of tool Simulation tool, based on Monte Carlo simulation, implementation unknown, maybe run on Prism
Availability of tools None. A description of the model is available at:
References Rosenthal, D.S., Rosenthal, D.C., Miller, E.L., Adams, I.F., Storer, M.W., Zadok, E., 2012. The economics of long-term digital storage, in: Memory of the World in the Digital Age Conference, Vancouver, BC. Retrieved from Rosenthal, D., Economic model of Storage, September 2011: Rosenthal, D., Economic model of Storage, November 2011:

The purpose of the model is to predict and compare the cost of long-term storage over time. The model predicts the costs of long-term storage for a data unit over a 100-year period. It covers the cost of capital of storage, and does not account for labour costs, presumably because labour cost are expected to be minimal for large amounts of data.

The model is not specific to any type of material as it focuses on storage. The model is a Monte Carlo simulation of the economic history of a unit of stored data. It has four high level components that model the costs of long-term storage:

  • Yield Curves
  • Loans
  • Assets
  • Technologies

Based on these four parameters the costs of storing one data unit is predicted and expressed as monetary costs.

This way of modelling storage cost is somewhat similar to known ways of continuously replacing equipment over time, but with a focus on uncertainty, both on the future interest rate used for discounting and on the future decrease in storage costs. Neither is truly exponential.

The model is based on up-front payment for long term (100 years) storage, not pay-as-you-go. The model uses discounted cash flow (DCF) to compare cost over time multiplying the interest rate by a short term factor and adds a planning horizon in years, which has to be paid off regardless of service life.

All storage technologies have a purchase cost, running cost, migration cost and a service time. New technologies arrive each year, and a technology incurs a purchase loan for the purchase cost and migration cost related to the previous technology, with a term equal the service life. Cost levels are based on experience from storage providers and a few large scale institutions.

After a few thousands run a Monte Carlo simulation can normally show what combinations fail, and what are durable.