Dr Xiaoxi Zhang

Postdoctoral Researcher
Department of Electrical &Computer Engineering

Carnegie Mellon University

Dr Xiaoxi Zhang is a Postdoctoral Researcher of the Department of Electrical and Computer Engineering at Carnegie Mellon University (CMU), working with Professor Carlee Joe-Wong. Before joining CMU, Xiaoxi received her PhD in Computer Science from The University of Hong Kong in 2017 under the supervision of Professor Chuan Wu and Professor Francis C.M. Lau. She attained her BE in Electronics and Information Engineering from Huazhong University of Science and Technology in 2013. Her research interests lie in the broad area of communication networks, including cloud computing systems, NFV systems, and wireless networks. The common theme of her work is a theoretical study of realistic networking problems, especially for optimized resource scheduling and network management. She is particularly interested in online algorithms, online machine learning, reinforcement learning, and algorithmic game theory. Her work involves optimization, algorithm design with theoretical analysis, and trace-driven simulations.

Dynamic Cloud Resource Provisioning and Pricing

On-demand resource provisioning in cloud computing provides tailor-made resource packages to meet users' demands. Public clouds nowadays provide more and more elaborated types of VMs, but have yet to offer the most flexible dynamic VM assembly. To fill this research gap, we investigate dynamic cloud resource provisioning and pricing in various problem settings. We first provide a novel truthful auction for VM provisioning and pricing. It has polynomial running time in expectation and near optimal social welfare, which show theoretical improvement compared with existing resource provisioning and pricing mechanisms. Next, we target a more realistic case of online VM auction design, considering arbitrary job arrivals and departures, as well as various server operational cost. We further extend the online auction to dynamic VM scaling with both scale-up and scale-out options for users, and to a stochastic online VM auction with both supply and allocation of resources. In addition, we also address cost minimization problems in two typical network service provisioning systems, Network Function Virtualization system and cloud Content Delivery Network. We novelly adopt online machine learning theories to estimate the uncertain inputs in the optimization problems. The predictions from online learning modules are further fed into novel online algorithms which can learn the user behavior and converge to the optimal decisions asymptotically. In particular, we deal with time-coupling decisions which can lead to an online dilemma and are not considered in classical online learning problems.