As the authors correctly suggest in the introduction, no code is perfect and selecting code for an application always involves tradeoffs. Although GRID codes--“a new family of erasure codes with high fault tolerance”--may not be optimal with respect to any single parameter, they may be the best codes for storage applications. This is because the combination of less-than-optimal parameters is better than any other class of codes.

The paper includes examples that show fault tolerance up to 15, efficiency up to 80 percent, very regular architectures, completely XOR-based operations, optimal small-write performance, and low reconstruction cost using local sets of data. Other codes exist that are better for any single feature. For example, Reed-Solomon codes have higher fault tolerance and optimal efficiency, but require more complex operations to detect and correct errors; this results in lower bandwidth for data transmission. Therefore, for storage applications, the higher bandwidth might be better than higher fault tolerance and higher efficiency.

The paper is very well written and accessible to nonspecialists. It compares and contrasts various classes of codes. It is an excellent paper for readers who want to learn the advantages and disadvantages of using a variety of codes.