Locally Repairable RapidRAID Systematic Codes — One simple convoluted way to get it all
Anwitaman Datta · 2014
The need to store humongous volumes of data has regurgitated the study of erasure codes, so that reliable fault-tolerant distributed (for scaling out) data stores can be built while keeping the overheads low. In the context of storage codes, one of the most vigorously researched aspect in the last half a decade or so is their repairability - which looks into mechanisms to rebuild the data at a new storage node, to substitute the loss of information when an existing node fails. Desirable (sometimes mutually conflicting or reinforcing) repairability properties include reduction in the volume of I/O operations, minimize bandwidth usage, fast repairs, reduction in the number of live nodes to be contacted to carry out a repair (repair locality), repairing multiple failures simultaneously, etc.