A method for distributed file transfers with high performance and reliability includes creating, on a first Trivial File Transfer Protocol (TFTP) server, a global cache, where the global cache is used to store up to a first portion of a data file. The method further includes storing, on the first TFTP server, in the global cache, one or more continuous data blocks that have exceeded a defined first request rate threshold, where the one or more continuous data blocks make up a subset of the data blocks of the first portion of the data file. The method further includes predicting, on the first TFTP server, a next data block in the data file to be stored in the global cache and in response to predicting the next data block, storing, on the first TFTP server, the next data block in the global cache.