David S. H. Rosenthal
Chief Scientist, LOCKSS Program
At the CNI Fall 2010 membership meeting, Serge Goldstein described Princeton’s POSE (Pay Once, Store Endlessly) service in which, based on the history of exponential decrease in storage costs, data is endowed with a capital sum to fund its storage indefinitely. Discussions sparked by this presentation, research at the Storage Systems Research Center at the University of California, Santa Cruz, and work for the Library of Congress on cloud storage for LOCKSS boxes, have all encountered questions whose answers require a more sophisticated approach to modeling the costs of long-term storage. Among these are:
- How can the cost of local storage, which has both capital and running costs, be compared with cloud storage, which has only running costs?
- Can solid-state storage, which is more expensive to buy but cheaper to run, be cost-effective for archival use?
- How vulnerable is the endowed data model to changes in the rate of decrease of storage costs?
Based on recent research into accounting for long-term costs, both practical from the Bank of England and theoretical from the Santa Fe Institute and Yale University, initial work supported by the Library of Congress is in progress to develop a Monte Carlo model of long-term storage costs. It consists of a framework into which models including interest rates and technology evolution, and policies including technology deployment and replication, can be plugged. The model can be used to explore a wide range of “what-if” scenarios.
The session will describe the model, present initial results and discuss future directions.