摘要 |
A method of operating a distributed object storage system includes selecting a set of storage elements 34, e.g. hard drives, upon each of which are stored a number of level-1 encoding fragments 730 of an object to be stored 520. The number of storage elements is equal to the basic number of storage devices 812 which are necessary to reconstruct the data object, plus the desired number of redundant storage devices 822. A second set of storage elements 36 is selected, being fewer than the basic number of storage elements of the level-1 storage set, upon which a group of level-2 fragments 750 is stored. The stored object may be reconstructed from the level one or level two fragments, or a combination of both. By encoding the object twice, additional redundancy can be obtained. The encoding of the fragments may be based on an erasure code. |