KOALA-C: A Scheduler for Integrated Multi-Cluster and Multi-Cloud Environments

The emergence of Cloud Computing offers users and companies the opportunity to utilize elastic resources without the burden of maintaining physical computing centers themselves. Among the popular cloud computing service paradigms today, the Infrastructure-as-a-Service (IaaS) provides virtual computing resources in term of virtual machines (VMs), which users can acquire and release according to their own demand. The pay-as-you-go pricing policy also makes it reasonable for both long-term and short-term usage. As a result, public clouds such as Amazon Web Service, Rackspace, and Microsoft Windows Azure have achieved great success in the cloud computing business. Cloud computing has also attracted much attention from companies and scientific communities who require large-scale computing power. Although work in the field of performance analysis and strategies of using elastic resources in clouds already exists, for institutes who already have their own resources, using clouds to extend their existing capacity is more desirable. This motivation derives a new heterogeneous environment, consisting of both clusters and clouds. Moreover, a deeper understanding of resource management in such an environment needs to be obtained. In this work, we propose the design of KOALA-C, a task scheduling system that can operate in integrated multi-cluster and multi-cloud environments. KOALA-C is an extension of two previous work: KOALA and SkyMark. KOALA-C establishes a universal view in resource management, which brings down the system complicity and makes it easy to manage both clusters and clouds. KOALA-C also keeps comprehensive track and statistics of the system for both analysis and performance evaluation purposes. Finally, KOALA-C has a flexible framework for designing scheduling and provisioning policies. With this framework, we design several policies for task scheduling and resource provisioning. These policies are evaluated through both trace-based simulations and real-world experiments using KOALA-C. The cost-performance trade-offs are analyzed and discussed.