Scaling OpenStack Clouds Using Peer-to-peer Technologies

Abstract

OpenStack is an open-source software platform for cloud computing, mostly deployed as an infrastructure-as-a-service (IaaS) and has a user base in industry and academia to date. Despite its popularity, OpenStack still has drawbacks in terms of scalability of number of compute nodes (metal machines) in a single cloud instance. More precisely, a single standard OpenStack cloud instance does not scale well and fails to handle user request once its number of compute nodes reaches a particular amount. The particular amount depends on how the cloud instance is deployed and how many computing resources are provisioned to the cloud instance. This thesis proposes a solution that allows to scale up OpenStack cloud instances by using peer-to-peer (P2P) technologies. The solution abstracts multiple OpenStack cloud instances as one, providing the same user experience as using a single and standard OpenStack cloud instance. This thesis was done at Ericsson Research Department in Stockholm, Sweden. In the thesis, we design and develop a proof-of-concept of the solution by implementing a software agent which runs on an OpenStack cloud instance, working as a message broker and providing OpenStack services to users. Association of agents is achieved by an inexpensive group membership protocol – Cyclon. We evaluate our P2P-based solution by comparing its system performance with a standard OpenStack deployment in terms of response time, failure resistance and CPU utilization. Results show that it is feasible to integrate virtual resources across multiple OpenStack cloud instances while abstracting them as a single cloud instance. Moreover, it is also shown that the proposed approach has higher failure resistance to certain operations (e.g. upload image and boot virtual machine). In addition, the solution has no limitation on a number of cloud instances and its performance, such as response time, failure resistance and CPU utilization, improves with the increasing number of cloud instances.