One-to-Many File Transfers Using Multipath-Multicast with Coding at Source
Ogawa Kyohei, Takeshi Iwamoto, Masato Tsuru · 2016
The rapid penetration of cloud and distributed computing technology is accelerating the need for one-to-many file transfers for sharing and duplicating data and applications among distributed data centers in an efficient manner. The purpose of our work is to develop a scheme to transfer a large-sized file from a single sender to multiple receivers over an OpenFlow-based network as quickly as possible. In Multipath-Multicast (MPMC) file transfer model we proposed previously, assuming that a file is appropriately divided into multiple equally sized blocks, a file (i.e., a set of blocks) is transferred to all receivers over multiple phases. In each phase, different blocks are concurrently transmitted to the same receiver over multiple paths to fully utilize the link capacities, while the same block is concurrently transmitted to multiple receivers by multicast to efficiently utilize the link capacities. In this paper, we propose an efficient schedule search procedure in MPMC to construct a good "schedule" that consists of the allocation of link capacities to flow routes from the sender to receivers and the allocation of blocks to those flow routes. Furthermore, to make the performance more stable and better when the number of trials in the schedule search is limited, we introduce the coded-MPMC scheme, in which the sender can generate a coded block by combining two or more plain blocks in an XOR coding manner and send it to multiple receivers. Through simulation on a large-scale topology, the schedule search procedure in MPMC model is validated to be applicable to a large network effectively and the coded-MPMC is shown to more likely find a better schedule than the original scheme that is not coding-enabled.