In this paper, we analyze the performance improvement of cloud systems by adopting nonvolatile memory. Specifically, we present the optimized adoption of non-volatile memory for the storage of cloud systems. Our analysis shows that a bulk of storage accesses does not happen on a single storage partition, but it is varied significantly for different cloud applications. In particular, journal accesses dominate in database applications, whereas swap accesses dominate in memoryintensive applications. In some applications such as video streaming, file accesses dominate. Based on these observations, we recommend the usage of non-volatile memory for maximizing the performance gain in cloud systems.