Dissertation Talk: Machine Learning for Resource Management in the Datacenter and the Cloud
Colloquium | May 10 | 1-2 p.m. | 465H Soda Hall
Neeraja J. Yadwadkar, UC Berkeley
Traditional resource management techniques that rely on simple heuristics often fail to achieve predictable performance in contemporary complex systems that span physical servers, virtual servers, private and/or public clouds. My research aims to bring the benefits of data-driven models to resource management of such complex systems. In my dissertation, I argue that the advancements in machine learning can be leveraged to manage and optimize todays systems by deriving actionable insights from the performance and utilization data these systems generate. To realize this vision of model-based resource management, we need to deal with the key challenges data-driven models raise: uncertainty in predictions, cost of training, and cost of updating the models.
In this talk, I will discuss these broad themes in the context of two problems: scheduling jobs on a cluster and virtual machine (VM) selection in the public cloud. I will begin by presenting Wrangler, a system that predicts when stragglers (slow-running tasks) are going to occur based on cluster resource utilization counters and makes scheduling decisions to avoid such situations. Wrangler introduces a notion of a confidence measure with these predictions to overcome modeling uncertainty. I will then describe our Multi-Task Learning formulations that share information between the various models, allowing us to significantly reduce the cost of training. Finally, I will present the highlights of our work on the PARIS system that enables cloud users to select the best VM (virtual machine) for their applications in the public cloud environments.