Introduction  Software  Hardware  Business/Social 

---

Hardware/Network Architecture

Long term digital data preservation is an inherently difficult task, because computer/network hardware tends to be relatively short lived (5-30 years) and usually becomes obsolete before it fails. Our approach to build a robust, long lived system is to create multiple dimensions of duplication and design for easy forward migration of components and architecture.

Techniques for Reliable Data Storage

Hardware Reliability

Duplication of data (symmetric redundancy)

RAID 5 per server

Two mirrored servers per location

Diversity of copies (asymmetric redundancy)

Media

Hard Drive

CD-ROM

Norsam disk

Location

San Francisco

Ely, Nevada

Additional (undisclosed) location

Rigorous Migration (Regular staged replacement of hardware)

Expansion of Mirror sites

---

Mirroring Architecture

---

Hardware Component Candidates

RAIDZONE sells the RS15-R1200 server that stores about 1TB, has dual CPUs running Linux, up to 1GB of RAM, and dual ethernet ports. Thus, we can likely store the live repository and run the server software on the same machine. Because it uses cost effective IDE hard drives, it only costs about $21K in a 1TB configuration.

JVC has a 100 disk archival CD-ROM jukebox, the MC-2102 for about $12K with an internal color printer. It can record about 80GB/day in even a minimal configuration, and looks like is a good choice for mass recovery of data as well as custom one-off backups (e.g. for a particular client).

Mitsui offers an printable, archival CD-R with a claimed lifespan of 100 years for less than $2 apiece in bulk (about $3K to backup the 1TB server)