Clouds have become an attractive infrastructure for high performance and scientific computing because the clouds offer cost efficiency, scalability, and elasticity of on-demand resources. Predictive resource management is developed to efficiently leverage cloud resources with two interrelated goals: ensuring application performance and minimizing execution cost. However, existing approaches are not sufficient to meet these two goals due to uncertainties in the clouds -- workload and performance uncertainties -- resulting in poor performance and adaptability in the cloud resource management.

My presentation introduces two techniques that mitigate such uncertainties for predictive resource management. I will first present a novel workload prediction framework called "CloudInsight" that leverages a combined power of multiple workload predictors. Next, I will focus on "Orchestra" framework, which ensures the performance goals of multiple cloud applications with dynamic allocation of shared resources in the user space. Then, I will conclude this talk with my vision for future research.