Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Yue Tan Dissertation.pdf (891.28 KB)

ETD Abstract Container

Abstract Header

Stochastic Modeling, Optimization and Data-Driven Adaptive Control with Applications in Cloud Computing and Cyber Security

Tan, Yue
http://rave.ohiolink.edu/etdc/view?acc_num=osu1431098853

Abstract Details

2015, Doctor of Philosophy, Ohio State University, Industrial and Systems Engineering.
Big Data has flown into every sector of the global economy ranging from social networks to online business to finance to medicine. With the rapid growth of data in many applications in the society, operations research (OR) professionals must shift to a broader view of developing analytical solutions characterized by the integrated use of data, processes and systems. Classical stochastic modeling, although proved to be useful in many traditional application areas (e.g. call centers, manufacturing systems), few works have been done in new applications arising from big data. Existing methods are lack of integration between data and modeling, recent development in adaptive control fails to address these new applications. In this dissertation, we aim to fill these gaps by developing new stochastic modeling, optimization and data-driven adaptive control approaches for managerial problems such as the resource provisioning of cloud computing and password management in cyber security systems. Resource provisioning of cloud computing, the task of planning sufficient amounts of resources to meet the service level agreements (SLAs) for all cloud users, has become an important management task in modern service clouds. We first present a stochastic modeling and optimization approach. We focus on the on-demand services and consider service availability as the key SLA constraint. We represent the problem as a capacity planning problem modeled by a multi-class Erlang loss network under service availability constraints. We show that the conventional square-root staffing provisioning solution which is based on Normal approximation, although asymptotically optimal, could lead to SLA violations with large probability. Instead, we propose using Poisson approximations and derive a tight upper bound for the Erlang loss probability, which would then yield new provisioning solutions. We show that our provisioning solutions are not only asymptotically exact but also provide better SLA performance. When the demand distribution is unknown, we adopt a stochastic gradient-based learning approach. Based on our stochastic loss model, we formulate the provisioning problem from a revenue management perspective. Our algorithm adaptively adjusts the provisioning solution as observations of the demand are continuously made. We prove that this algorithm converges to optimum and adapts to non-stationary demand. We then extend our data-driven adaptive control framework to password management in cyber security systems. A password policy is the frontline of protection against cyber attacks, which contains a set of rules on password length, duration, etc. We assume password has censored lifetime, and policy maker determines the duration of the password without complete knowledge of its true lifetime distribution. We develop a gradient based algorithm integrated with a Bayesian learning framework. We show that our algorithm converges to optimal solution and adapts to non-stationary lifetime data. While the problems we consider are in the context of cloud computing and cyber security, the models and methodologies we developed are important from the theoretical perspectives of investigating 1) stochastic loss networks and Poisson approximations, and 2) Bayesian learning and stochastic gradient methods for censored observations, and of investigating the corresponding optimal resource allocation and data-driven adaptive control problems. Our analysis and results can also support a wide range of practical problems in various application domains.
Cathy Xia (Advisor)
Theodore Allen (Committee Member)
Guzin Bayraksan (Committee Member)
134 p.
Operations Research

Recommended Citations

Citations

  • Tan, Y. (2015). Stochastic Modeling, Optimization and Data-Driven Adaptive Control with Applications in Cloud Computing and Cyber Security [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1431098853

    APA Style (7th edition)

  • Tan, Yue. Stochastic Modeling, Optimization and Data-Driven Adaptive Control with Applications in Cloud Computing and Cyber Security. 2015. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1431098853.

    MLA Style (8th edition)

  • Tan, Yue. "Stochastic Modeling, Optimization and Data-Driven Adaptive Control with Applications in Cloud Computing and Cyber Security." Doctoral dissertation, Ohio State University, 2015. http://rave.ohiolink.edu/etdc/view?acc_num=osu1431098853

    Chicago Manual of Style (17th edition)

osu1431098853
453
© 2015, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.