On the impact of erasure coding parameters to the reliability of distributed brick storage systems

For a given amount of storage overhead, erasure coding offers a higher degree of survivability than pure replication. Consequently, erasure coding attracts much attention these years in the research area of reliable distributed storage systems. Although numerous erasure codes have been put forward, how to choose erasure coding parameters to maximize system reliability has not yet been sufficiently investigated. Erasure coding parameters greatly affect not only system survivability with multiple concurrent failures, but also data repair speed. We propose a method that can quantitatively evaluate these effects. Based on the method, the issue of determining erasure coding parameters is solved. Besides, relationships among other reliability-affecting factors such as storage overhead, repair bandwidth and single brick´s properties are also investigated.