Though erasure codes are widely adopted in high fault tolerance storage systems, there exists a serious small-write problem. Many algorithms are proposed to improve small-write performance in RAID systems, without considering the network bandwidth usage. However, the network bandwidth is expensive in cloud systems. In this thesis, we proposed an ECL-based (E-MBR codes, Caching and Logging-based) technique to improve the small-write performance without using extra network bandwidth. In addition, the ECL-based technique also reduces the delayed parity update and data recovery latency compared with the competing algorithm.