VaaS: Visualization as a Service

Traditional HPC visualization applications provide scientists with scalable solutions to work with large datasets. However, computation today is moving towards cloud-hosted systems, as they present key advantages over traditional HPC cluster systems, primarily that of high availability. Computation on a cloud-hosted system can be accessed on demand without long queue times and without long resource acquisition times. Existing applications cannot be ported to the cloud as they rely on a monolithic application structure that does not flexibly distribute among virtual cloud instances.In this dissertation, I study the architectural decisions required to transition scientific data analysis and visualization into the cloud. In particular, I focus on a microservice-based architecture model, designed to use loosely couple modular components using a limited communication model. This general scheme allows scalability and flexibility on cloud-hosted systems, and provides Visualization as a Service that can be deployed on-demand. I present working prototype solutions to the challenges of large data storage and working set management, loosely-coupled rendering and analysis services, and visualization records keeping and reproducibility.