Octopus: Efficient Data Intensive Computing on Virtualized Datacenters

Svitlana Tumanova, Olga Irzak, Lili Sun, Shiri Margel, Eyal de Lara

6th International Systems and Storage Conference, Haifa, Israel, June 2013



Virtualization provides many bene fits for running data intensive workloads including security, performance isolation, and ease of management and confi guration. Unfortunately, current VM technology prevents taking advantage of sharing opportunities, resulting in substantial network traffic and application slowdown. Octopus is a new framework for running data intensive applications on virtualized datacenters. Octopus provides efficient file sharing across VMs running on the same physical host and optimizes the placement of VMs in the cluster to maximize sharing opportunities. Our experiments with a suite of bioinformatics and natural language processing applications show that Octopus reduces network transfer by up to 83% and total runtime by up to 55%