Compared to 3-way replication, which incurs triple storage costs for data availability and reliability, erasure coding has emerged as a good alternative to ensure data safety with significantly lower storage overhead for large-scale cloud storage systems. However, this results in high degraded read latency and long reconstruction time owing to higher disk and network traffic where unavailable data are erasure coded. In this paper, we aim to improve the recovery performance for coded data when failure occurs, by shortening the reconstruction time and degraded read latency. Two optimization methods are proposed: (a) proactive data migration using drive failure prediction, (b) reactive recovery optimization when failures actually occur. This paper includes the current research direction and the future work that the author aims to perform in the next years. These results show that the reconstruction time of unavailable data could be reduced by these optimizations.