Coding Overhead Analysis of Decentralized Coded Caching

Lei Zheng, Chao Tian, Qingchun Chen

Abstract: In this letter, we investigate the coding overhead in decentralized coded caching scheme. Firstly, a unified framework is presented to analyze the incurred coding overhead for two header formats. Secondly, the coding overhead analysis of three decentralized schemes, including the Maddah-Ali-Niesen (MAN) scheme (Maddah-Ali and Niesen, 2015), the coded prefetching scheme (Tian and Chen, 2018), and the reduced subpacketization scheme (Jin et al. , 2016), is presented to unveil that, the overhead can lead to performance loss. Our analysis shows that the second format is always preferable for any uncoded and binary coded packets, and the coding overhead increases with the increase in subpacketization, while the degradation rate of transmission payload decreases exponentially. It is demonstrated numerically that, for the system with small number of users and moderate sized files, the coding overhead tends to be negligible in the uncoded and binary coded schemes; however, it becomes significant for non-binary coded schemes, especially when the file size is not sufficiently large. These results suggest the importance of choosing an efficient header format in specific coded caching application, in particular for non-binary schemes.