The invention provides a cloud storage performance optimization method and system, and relates to the technical field of the Internet. The method comprises the steps of: acquiring partition information of each partition of a storage node in a Swift cloud storage architecture; determining a first partition and a second partition in each partition based on the partition information; and migrating files from the first partition to a proxy node and migrating files from the second partition out of the proxy node. The purpose of improving the access speed of a cloud storage system is achieved by dynamic migration of hotspot data, thereby solving the technical problem that the traditional cache strategy in the prior art cannot effectively improve the read-write speed of mass small files.