Supporting software-defined HA clusters on OpenStack platform

Server virtualization is one of the critical cloud computing technologies since it enables the use of software-based virtual machines (VMs). Many enterprises have already adopted this technology to build their cloud platforms because of its scalability and flexibility. However, high availability (HA) remains a challenging issue on such a virtualized cloud platform. That is, the time for VM failover must be short to reduce possible loss of business and data. For an organization with many independent departments, the cloud is usually partitioned into several clusters of physical machines, and the recovered VMs from failures need to be placed inside its departmental boundary. To this end, we propose a new mechanism that enables software-defined-HA clusters over a popular cloud platform, OpenStack. The proposed mechanism can detect various types of failures, including OpenStack Nova compute service failure, Libvirt (cloud middleware on each physical machine) failure, host OS failure, and physical machine failure. With the new mechanism, the cloud administrators can partition the OpenStack compute pool into several software-defined-HA clusters, on which a VM executes and failovers inside its cluster. Once the number of physical machines goes down to one or zero, the HA cluster is then labelled unprotected. To enable restricted failover, the administrators need to associate the VM users with a HA cluster. The VM users are then able to protect/unprotect VMs is in a transparent way.