Jiang Haibo

Improved decoding algorithm for the generalized EVENODD array code

Data placement scheme based on the generalized EVENODD array code was one of the most important choices for data storage systems to tolerate failures. The efficiency of the erasure code was directly related to the number of XOR operations during the encoding and decoding. We provided an improved decoding algorithm of the generalized EVENODD array code for correcting various triple node failures. It showed that the decoding complexity of the new algorithm was much lower than those of the existing comparable algorithm, thus it made the generalized EVENODD array code more efficiency in storage systems.

Short-length rateless codes with unequal error protection property

In some special applications, the importance of different parts of a file to be transmitted such as video or picture is different and the important part (such as the format part) need be given prior protection. This is called unequal error protection (UEP). In this paper we propose a new coding scheme which can provide UEP by cascading new short-length rateless codes with RS codes. In our proposed UEP coding scheme, the decoding success probability of the important data parts are improved by participating in two encoding processes. We theoretically analyzed the decoding failure bound of our UEP coding scheme and confirmed the effectiveness of our method by numerical experiments. Experimental results showed that even the code length is small, our UEP coding scheme can also realize unequal error protection under very small decoding overhead.

A new class of rateless codes

Classic rateless codes, such as Raptor codes and LT codes are long codes. But special applications such as multimedia broadcasting and Multicast Server usually require small length codes (a few hundred in size). In response to this demand, we propose a new construction method of small length rateless system codes. The generator matrix of our proposed coding scheme is a special 0–1 random matrix in which every element is valued independently and identically and has same probability to be valued 1. Theoretical analysis and simulation experiments show that, when using maximum likelihood decoding, the rateless code generated by our method can be decoded with extremely high probability under the limitation of both small code length and small decoding overhead.

A new storage scheme for streaming media server cluster

We report on the design, implementation of a non-equal scheme in streaming media data storage. According to the characteristics of the streaming media and considering the human beings usage, our scheme combines n-way mirroring and erasure-coding together to storage streaming media data. It is make good tradeoffs between system fault-tolerance and sufficient quality of service streams. The system has the optimal storage efficiency and the remarkably fault-tolerant capability. At last the design of the non-equal streaming media data cluster storage system is introduced in detail, and an intensive performance evaluation study on it is performed.