Dynamic load-balanced multicast based on the Eucalyptus open-source cloud-computing system

In clouds, many applications need to distribute large data sets from the cloud´s storage facility to all compute nodes as fast as possible, especially data-intensive parallel applications. Many multicast algorithms have been used for clusters and grid environments. In order to maximize available bandwidth and avoid bottleneck links, a common approach is to construct one or more spanning trees based on the network monitoring data and network topology. However, in clouds the available bandwidth changes dynamically, so delivering optimal performance becomes difficult. In this paper, we focus on Eucalyptus (an open-source cloud-computing platform) and propose a high performance multicast algorithms `steal-and-p2p´ based on `non-steal´ and `steal´ algorithm mentioned in [1]. We evaluate our algorithm on Eucalyptus, and show that the algorithm can achieve high throughput and perform much better having each node downloading all data directly from storage facility.