Vault : an open-source parallel database as a service

Abstract

In recent years, parallel processing technology has become remarkably popular for enterprise big data analytics. However, the traditional IT infrastructure sets a barrier to enterprise big data analytics because of its limitations on scalability and high total cost of ownership. Migrating the parallel database system to the cloud platform offers enterprises scale up or down on demand without consideration of on-site hardware investment (e.g., on-site hardware maintenance and repair). Besides, the multi-tenancy property in a cloud platform can minimise total cost of ownership by sharing the parallel database system among multiple tenants. This thesis presents Vault, an open source cloud-based service which aims to provide parallel database-as-a-service (PDaaS) at a low operational cost with the service level agreement, SLA (i.e., a commitment governing the minimal level of service agreed between a service provider and tenants). Vault is built on top of the cloud platform, OpenStack, which is an open-source software that offers a cloud infrastructure for the parallel database system to carry out data analytics. With the advent of resource sharing in the multi-tenant environment, the service provider gains advantages of maximising resource utilisation and minimising the operational cost. Our experiments present that Vault serves tenants with only 55.2% of the requested nodes in OpenStack cloud while a 99% query-latency SLA is still guaranteed with high availability.