Optimal Fraction Repetition Codes for Access-Balancing in Distributed Storage

To solve the access-balancing problem in distributed storage systems, we introduce a new combinatorial model, called MinVar model for fractional repetition (FR) codes. Since FR codes are based on graphs or set systems, our MinVar model is characterized by the property that the variance among the sums of block-labels incident to a fixed vertex is minimized. This characterization is different from Dau and Milenkovic's MaxMinSum model, while the minimum sum of labels is maximized. We show that our MinVar model is meaningful by distinguishing labelings with different variances but with the same MaxMin value for some FR codes. By reformulating the MinVar model to an equivalent vertex-labeling problem of graphs, we find several families of optimal FR codes with balanced access frequency, and provide fundamental results for both problems. It is interesting that MinVar model is closely related to the concept of magic-labeling in graph theory.
Comment: accepted to IEEE Transaction on Information Theory