Combinatorial and Rectangular Layouts as Building Blocks for Local Reconstruction Codes

Modern data centers protect the contents of their data using erasure correcting codes. In recent years, locally repairable codes have been proposed that allow dealing with the most frequent case, a single unavailable disk or a single unreadable sector by only using redundant information in the rack where the failure has been detected. Recently, Pâris proposed a simple method for creating locally repairable layouts where several local layout in different racks are “bundled” together by creating additional inter-rack reliability stripes that provide protection against rack failure and offer additional opportunities for recovery if a local layout fails in its protection task. We propose here a flat rack-internal layout that is more dispersed than a rectangular layout that also provides fast recovery for single failures. The research question is to what extent dispersion improves reliability. As we will see by example, dispersion does provide better robustness, but fails to significantly improve the five year probability that inter-rack codes have not to be used.