From DBWiki

N-way Redundancy

N-way redundancy ensures that a file is available as long as one storage peer is remaining on the network. As shown in the image above, file-A.dat would be considered N-way redundant. If needed, Starfish allows the redundancy of data to be specified on a file by file basis.

N-way redundancy is the best method for ensuring that data is available at all times, regardless of sub-system failure. It is a method of providing fault-tolerance by saturating the entire system with multiple copies of a piece of data.

For an extreme example of this concept, consider the following:

In a 20-node Starfish storage system where you wanted to ensure that a piece of data was accessible at all times, N-way redundancy requires that you copy that file to all 20 nodes. By providing every storage peer with a copy of the file, you ensure that even if 19 nodes fail - the remaining node will be able to serve the file to any file system client that requests it.

The redundancy level of files on Starfish is determined on a file-by-file basis by the person that stored the information. A file has an extended attribute associated with it that can specify how many mirrors of the file should exist on the file system at any given point. If one of the mirrors fail permanently, Starfish ensures that the file is mirrored to another storage peer.

Note that there are two parts that are required to ensure that N-way redundancy works. The first is that the file system keeps functioning after multiple storage node failures. This is called high availability and it is currently implemented and functioning. The second is called mirroring, and that is currently not implemented. The plan is to have it implemented and launched by late August 2008. Check the Downloads page for more information on features in recent releases.