Resource Management of Virtual Infrastructure for On-demand SaaS Services

With the emerging of cloud computing, offering software as a Service appears to be an opportunity for software vendors. Indeed, using an on-demand model of provisioning service can improve their competitiveness through an invoicing tailored to customer needs. Virtualization has greatly assisted the emerging of on-demand based cloud platforms. Up until now, despite the huge number of projects around cloud platforms such as Infrastructure-as-a-Service, less open research activities around SaaS platforms have been carried on. This is the reason why our contribution in this work is to design an open framework that enables the implementation of on-demand SaaS clouds over a high-performance computing cluster. We have first focused on the framework design and from that have proposed an architecture that relies on a virtual infrastructure manager named OpenNebula. OpenNebula permits to deal with virtual machines life-cycle management, and is especially useful on large scale infrastructures such as clusters and grids. The work being a part of an industrial project1, we have then considered a case where the cluster is shared among several applications owned by distinct software providers. After studying in a previous work how to implement the sharing of an infrastructure in such a context, we now propose policies and algorithms for scheduling jobs. In order to evaluate the framework, we have evaluated a prototype experimentally simulating various workload scenarios. Results have shown its ability to achieve the expected goals, while being reliable, robust and efficient.