Analytical Modeling of Reconstruction Delays in Distributed Storage Data Centers

摘要

Cloud users store millions of files on the cloud every day. Often times the files are simply replicated so that if one is lost a mirror can be fetched when required. Simple replication of data files is easier however it is more costly since more storage is required. Alternatively, cloud datacenters use coding techniques that code a file into parts; each of which can be stored at a different location and sometimes geographically distant locations. The biggest advantage of coding is that not all the parts are required to reconstruct the file in case one of more parts are lost. This is especially useful in presence of disk failures that frequently occur in datacenters. In this paper we present an analytical model of the reconstruction process with reference to delays incurred during the reconstruction process.