[PAST EVENT] Dynamic cloud resource allocation and pricing
Speaker: Dr. Xiaoxi Zhang, Carnegie Mellon University
Title: Dynamic cloud resource allocation and pricing
With the proliferation of cloud computing, users can purchase computing resources from cloud platforms such as Amazon EC2 and Google Cloud Platform. Online resource allocation and pricing lies in the center of maximizing profit for the cloud provider while satisfying users? Quality-of-Service requirements. However, under today?s conventional pricing scheme, the pay-as-you-go plan with fixed prices on pre-determined resource configurations, neither users are able to maximize their utilities or the provider can maximize her profit. Aiming to improve the current cloud services, we propose new allocation and pricing schemes with provable performance guarantees. While we adopt auctions and posted price mechanisms, similar to Amazon?s Spot pricing, our schemes guarantee no interruptions on acquired resources for users, providing more stable services than Spot Instances. In this talk, I?ll first show our allocation and pricing mechanisms designed for cloud providers, and then introduce an optimal cost-effective bidding strategy for users who launch distributed machine learning jobs on Spot Instances, accounting for the effects of job interruptions on cost, training error, and completion time.
Dr. Xiaoxi Zhang is currently a Postdoctoral Research Associate at Carnegie Mellon University. Before joining CMU, she received her Ph.D. in Computer Science from The University of Hong Kong in 2017 and her B.E. in Electronics and Information Engineering from Huazhong University of Science and Technology in 2013. Her research interests include cloud computing, optimization, online algorithms, machine learning, distributed systems, and network economics.