
Perennial archives.
Briefly, this means we provide the resources to overcomes the technological 
obsolescence of supports and drives we use.

1. The perennial archival comes with 3 main objectives:
- conserve the document
- make it accessible
- preserve it's understanding

1.1. Supports reliability.
On numeric ages, time life of hard drive and optical supports is 5 years.
The numerical archives on hard drives or tapes are sensitive to the earth's 
magnetic field. Burning WORM supports and store them in good conditions 
provides a compromise (norme NF-Z-42-013). Persons prefer do not export 
data on support and let them on redundant servers, but is-it responsible ?

1.2. Identify
The document understanding shall be linked to other documents defining a 
context. Consequently, it is important to provide a way to identify 
references from documents to others. Grouping documents on collection make 
index easier and provide a means to determine the scale of data to preserve.

1.3. Another data
Just as supports need a compatible drive, crude data need software to read 
them. It is important to memorise the software and format versions, if not 
possible store software's source codes and/or format specifications. In 
order to do so, we need meta-data.

2. Geographical duplication
In order to mitigate natural and technological disasters, archives must be 
duplicated on several sites.

2.1 Technical solutions
Computer science offer several means for redundancy but finally, the 
meta-data management will remains centralised. The NICT mainly based on 
databases are using "hight availability" to offer crash recovering. Is-it 
the real purpose? Programmers use revision control system to share their 
source codes. Although using a centralised deposit, this solution offer the 
advantage to desynchronise updates, that is a way to anticipate crash 
recoveries.

2.2 Distribute meta-data as source code.
Using a revision control system is not so obvious:
- Meta-data files must be split into acceptable size so as to be merged in 
memory.
- Meta-data must be human readable as automatic merge may fails.
- We need enough memory to load Meta-data.
Nevertheless, revision control system not only provides an elegant solution 
for geographical duplications and crash recovery, but also the change 
history.

3. Archive reversibility.
A significant effort will be require to extract archives in order to return 
them:
- need to compact archives to send them using a media (support or network)
- need to extract and export meta-data consequently
However, we shall consider that the ERMS becomes obsolete itself and we 
need to change it. Archiving on supports make sense here as we just have to 
bring them back. Parsers designed to load the meta-data offer a software 
library to export them into the wanted format.
