Communication (or data shuffling) is a major performance bottleneck in large-scale distributed computing frameworks for big data analytics. To alleviate this bottleneck, we propose “coded distributed computing” (CDC), which achieves the information-theoretic optimal tradeoff between the load of computation and the amount of required communication in a general class of distributed computing frameworks. CDC utilizes redundant computations to produce structured side information across computing servers, which then enables coding opportunities to significantly reduce the load of communication. We have empirically demonstrated the impact of CDC in various scenarios, and have shown that, compared with state-of-the-art uncoded schemes, CDC provides more than 3x speed-up in basic computations underlying big data analytics applications. Furthermore, CDC can also be applied to mobile edge computing systems to break the scalability bottleneck due to limited bandwidth. It can also be utilized to mitigate the straggler bottleneck, achieving another tradeoff between the computation latency caused by the stragglers and the communication load. After illustrating these results, I will conclude the talk by discussing several exciting future research directions in this area.