a funding and operational model for long-term preservation and sharing of research data 1 educause...

A Funding and Operational Model for Long-Term Preservation and Sharing of

Research Data

1EDUCAUSE Live!, 9/1/2010

Special Thanks To

• Mark Ratliff• Digital Repository Architect

• Implemented DataSpace at Princeton

• Princeton University Library and OIT• Marvin Bielawski

• Joyce Bell (Metadata) and Dan Santamaria (archives)

• http://arks.princeton.edu/ark:/88435/dsp01w6634361k


Goal of DataSpace Model

• Store data “forever”• Preserve and protect

• Share data with everyone• Make available

• Make accessible

• Make it Work!• Financially

• Operationally

EDUCAUSE Live!, 9/1/2010 3

What is “Forever”?• Longer than a typical project?• Longer than a typical career?• Longer than a typical institution?• 5 years, 10 years, 25 years, 100 years?• Suggestion: treat data same way library treats

books• Intent is to preserve indefinitely• As long as practical, feasible• Cannot be precisely defined


Why Save Data “Forever”

• Because we want to make it:• Available to ourselves and our students and

colleagues• Where are the data sitting today? On a

departmental server? On a computer under your desk? On a CD or DVD somewhere?

• Where is your dissertation data?

• Available to future scholars, including ourselves



• Because we need to:• Encourage honesty?

• Gregor Mendel probably cheated

• Like open-source: help uncover mistakes, bugs?

• Open Data Movement• Mostly library/catalog data, map data, WordNet

• Open Access Movement• Mostly publications

• Because it’s not “our” data



Because we have to:

Science Insider, May 5 reports:

” Edward Seidel, acting head of NSF's mathematics and physical sciences directorate, described NSF's intention to require all applicants to submit a data management plan along with their grant application in a presentation this morning to the National Science Board, NSF's oversight body. …NSF's current policy requires grantees to share their data within a reasonable length of time so long as the cost is modest. "That's nice, but it doesn't have much teeth," said Seidel. Under the new policy, which is expected to be unveiled this fall, a researcher would submit a data management plan as a two-page supplement to any regular grant proposal. That would make it an element of the merit review process.”


Data Sharing Policies

• New NSF policy continues trend • NIH Data Sharing Policy (2003):

http://grants.nih.gov/grants/guide/notice-files/NOT-OD-03-032.html

“all investigator-initiated applications with direct costs greater than $500,000 in any single year will be expected to address data sharing in their application.”


NIH Data Sharing Policy

• “Applicants may request funds for data sharing and archiving. The financial issues should be addressed in the budget section of the application.”

• Specifics depend on grant, published in RFP, RFA or PA


NSF Data Archiving Policy

• Division of Social and Economic Scienes• http://www.nsf.gov/sbe/ses/common/archive.jsp• “Grantees from all fields will develop and

submit specific plans to share materials collected with NSF support, except where this is inappropriate or impossible.”


NSF Data Archiving

• From Grant Proposal Guide• NSF “expects PIs to share with other

researchers, at no more than incremental cost and within a reasonable time, the data, samples, physical collections and other supporting materials created or gathered in the course of the work.”

• Specifics depend on grant and program officer


Other agency Policies

• See Gary King’s Page on “Data Sharing and Replication”

• http://gking.harvard.edu/replication.shtml• See National Academy of Sciences “Ensuring

the Integrity, Accessibility, and Stewardship of Research Data in the Digital Age”, July, 2009

• http://www.nap.edu/catalog/12615.html


Current Storage Models

• Let someone else do it• Government agency/lab/bureau

• NOAA National Geophysical Data Center• GenBank (DNA data)• fMRIDC (fMRI publications and data)• NCSA Astronomy Digital Image Library


Current Storage Models• Professional society/Journals

• Global Ocean Observing System: coordinates distributed data

• Dryad: ecology/evolutionary biology

• Nice folks at another University• ICPSR, University of Michigan (political/social)• Dryad: ecology/evolutionary biology• Protein Data Bank (PDB): 3-D protein data• NCSA Astronomical Image Library

• The “Cloud”


Current Funding Models

• Institution/department pays• Grants pay monthly/yearly• Haphazard

• Some grant money

• Some departmental money

• Use whatever is available

• Don’t worry, someone will pay


Current Funding Models

• Most require some form of on-going payment• Advantages

• Capitalist approach to data storage

• If someone wants to pay, data gets saved

• “Natural” expiration process

• Disadvantages• Capitalist approach to data storage

• Who pays to save rarely used data?


Different Approach

PAY ONCE, STORE ENDLESSLY (POSE)

Why Pay Once?• Grants expire often and quickly• Researchers retire/move/move-on pretty often

How Store Forever?• Administrators expire slowly • Institutions expire rarely• Cost is FINITE (how is that possible?)

•Magic of Math


The Business Model (1)• I = Initial cost of storage

• D = rate at which storage costs decrease yearly, expressed as a fraction (e.g., 20% would be 0.2)

• R = How often, in years, storage is replaced

• T= Cost to store the data “forever”

T = I + (1-d)r * I + (1-d)2r * I + ….

If d=20%, r = 4:

T = I + (.84 )* I + (.88)* I + ….


The Business Model (2)

If d >0,

T = I + (1-d)r * I + (1-d)2r * I + …. CONVERGES

= I * 1/(1-(1-d)r)

For d=20%, r = 3: T=I * (1/.83): T ~= I * 2


Charge 2x initial storage cost, save half, store forever!

To simplify: Let S = 1/(1-(1-d) r) : “Storage Factor”, “S factor”

Then: T = I * S : Total Cost = Initial Cost * S Factor

“S” Factor is very stable


• “S” factor small, between 1 and 2 for broad range of reasonable values:

Rate of Decrease in Storage

• Is 20%/year too much?• 1981: 10 meg drive costs $3,000

• 2010: 500 gig drive costs $600

• 250,000 fold decrease over 30 years, or an average of 35% per year.

• 2000: IBM 20 gig drive costs $280

• 2010: 500 gig drive costs $600

• 12-fold decrease over 10 years, or about 23%/year.


An Example: DataSpace at Princeton

• FC costs decrease by about 16% per year• SATA costs decrease by about 17% per year


The “S” factor for Princeton

• SATA cost = $1.81/gb• Replace every four years• Costs decrease by 20% year

“S” = $1.81/(1-.8 **4) = $3/gb

Adding tape backup jumps this to $6/gb

$6K one-time to store a terabyte forever

$6K one-time to store a terabyte forever23EDUCAUSE Live!, 9/1/2010

But what about Other Costs?

• Cost of disk drives only (small) part of cost of saving/sharing data

• What about costs of people, buildings, electronics, software

• Some claim these costs “swamp” the actual cost of the disks


Model can handle other costs

• People, building, infra-structure, software costs MUCH higher than disk costs ….

• … but NOT if pro-rated across storage units.• Storage administrator in 2010 can manage 1000

times more storage than in 2000.• Have to look at “marginal costs”; how much

more in people costs to store an extra terabyte?


Model can handle other costs

• On a per unit of storage basis, people and electronics and software and infra-structure costs also decreasing rapidly

• These costs can be built in to the “S” factor (just as we built the tape backup costs into the “S” factor at Princeton).

• As long as a cost is decreasing over time on a pro-rated basis, model applies.


When does model break down?

• If costs increase on a per-unit basis over time, then you must have on-going income stream

• Administrative, data-translation, data-delivery costs are candidates

• Need to minimize such costs• Keep administrative overhead and service

overhead to a minimum


Organizational facet of model

• So far, only looked at financial facet• Financial facet needs to be combined with

organizational facet to keep costs down• Write Once, Read Forever (WORF)

• Set of principles aimed at minimizing operational costs

• Also implements the “sharing” requirement of current grant policies


WORF Principles

• Storage may not be “re-used”; it is archival, not working

• Permanent URL assigned; data cannot be changed. Can be made unavailable or super-ceded by new data.

• All data permanently publicly accessible (minus “embargo” period)

• Repository only provides storage and web access


WORF Principles (2)

• POSF• Ancillary services (data conversion) separately

charged

We propose that a repository implementing the above set of principles be called a “DataSpace Repository”


DataSpace at Princeton

• OIT and the Library have implemented a DataSpace repository

• Uses “Dspace” repository software developed at MIT and the Archival Resource Key (ARK) system for permanent URLs

• Just started:• http://dataspace.princeton.edu


a funding and operational model for long-term preservation and sharing of research data 1 educause...

Documents

data foreverbecause

map data

protectshare data

research data

dissertation data

data management plan

trend nih data sharing

nsfs current policy