On the Performance-Cost Tradeoff for Workflow Scheduling in Hybrid Clouds

The use of public clouds to extend the capacity of private resources has become a popular manner to achieve elasticity in the available in-house computational power to meet deadlines. Schedulers for such hybrid clouds have the role of deciding which types of instances should be leased in a pay-per-use basis to fulfill application demands. Often these schedulers assume a costless private cloud, which may not be a real scenario: aggregated costs can come from maintenance, energy consumption, administrative staff, and, more recently, from leasing a private data center from hardware manufacturers or a virtual private cloud from public providers. Based on these more realistic assumptions, we assess the behavior of a cloud scheduling algorithm in the face of different pricing relation between private and public resources. We show there exist a trade-off between using local resources to maximize the utilization of private cloud to minimize the monetary cost with outsourced resources and using leased resources from public cloud to satisfy deadlines, which can lead to idleness in the private cloud. Preliminary results indicate that application deadlines and the cost of private cloud can influence in the private cloud utilization, sometimes outsourcing most executions to public clouds, which leads to increased costs to run the application. We then argue that utilization-aware schedulers are also important to be developed when considering hybrid clouds.