Novel client-cloud architecture for scalable instance-intensive workflow systems

Though workflow technology is relatively mature and has been one of the most popular components of process aware systems over the last two decades, few workflow architectures can efficiently support a large number of concurrent workflow instances, i.e. instance-intensive workflows. The basic requirements include high throughput, elastic scalability, and cost-effectiveness. This paper proposes a novel client-cloud architecture which takes advantages of cloud computing to support instance-intensive workflows, presents an application level real-time resource utilization estimation model, and identifies two primary principles to ensure the sustainable scalability, namely: (1) the time for a load balancer checking must be less than the decaying time of a server instance when it is overloaded, (2) the sampling time for an alarming service plus the launching time of new server instance must be less than the decaying time of a server instance when it is overloaded. Based on the above, we design and implement the SwinFlow-Cloud prototype. Finally, we deploy and evaluate the prototype on Amazon Web Services cloud. The results show that the prototype is able to satisfy all the basic requirements for instance-intensive workflows.