On Enhancing the Performance of Inter-Cloud Data Transfers

The research presented in this paper concerns the transfer of data in the form of requests from a compute cloud to a storage cloud using a buffer at the compute system. Different buffering techniques from literature have been shown to improve performance either when the system is subjected to a high load or a low load, but not both. This paper discusses the performance of a threshold based and a machine learning based adaptive data transfer methods that dynamically change the buffering technique at runtime to offer an improvement in performance at all loads. The performance results of a proof-of-concept prototype are discussed for both the threshold based and machine learning based adaptive methods.