Dynamic Parallel Downloading with Network Coding in $\lambda$-Grid Networks

Abstract

In $\lambda$-grid networks, data files for job execution are stored on file servers as replicas, and computing servers, which execute jobs, download these replicas in parallel to reduce downloading time. However, parallel downloading raises blocking probability of lightpath establishments because it uses many links and thus wavelength resources are wasted. To resolve this problem, we propose a dynamic parallel downloading scheme with network coding which encodes data at intermediate nodes. The proposed scheme performs network coding by regarding file servers as intermediate nodes. In this scheme, a file is divided into multiple blocks.  A file server creates an encoded block from those blocks and stores it as a replica. Computing servers download encoded blocks from multiple file servers in parallel. Through simulation experiments, we show that the proposed scheme can improve the blocking probability and the downloading time efficiently.