Abstarct: Due to significant growth of data and predictions of this trend continuity, offering appropriate approaches for data storage is more necessary than before. By getting started of cloud storage systems in 2006, developing and improvement of this systems is highly considered. In this thesis, we improve the performance of cloud storage by representing novel methods. First, by exploiting new criterion “Message Importance Measure” that is used in minority subset detection, we compress data before outsourcing on cloud. With comparison of this method with Huffman, Shannon and Shannon Fano Elias, in cases that important part of message is considered, we can reduce storage space up to 30 percent. Also we investigate effect of symbols removing thresholds in this compression technique. In second part of thesis, we implement coding-baxsed cloud storage with LT codes and in order to improve performance of this systems in data retrieval, we use Shokrollahi, PRSD, CPRSD degree distributions in addition to RSD. By applying new distributions, we demonstrate that by applying new distributions, successful retrieval probability could be increased with fewer overhead. Moreover, we reduce retrieval time by using an algorithm baxsed on average and standard deviation of needed encoding symbols instead of original symbols in decoding process.