Oct 14, 2016
Lopata Hall, Room 101
"Pricing and Bidding Strategies for Cloud Computing Spot Instances"
Advisor: Roch Guérin
Cloud computing has experienced explosive growth in recent years, with numerous companies, e.g., Amazon, Aruba, Google, IBM, Microsoft, Salesforce, SAP, etc., offering a wide range of cloud services. Using Amazon EC2 as an example, it currently offers three main pricing options, i.e., on-demand, reserved, and spot instances, that provide different trade-offs between the level of commitment the user is willing to make and the cost of the service.
This talk is concerned with a cloud computing offering based on the spot service option, and focuses on exploring how a cloud provider should set spot prices given a user population with heterogeneous profiles, i.e., job valuations and sensitivity to delay. We show that under reasonable assumptions users that decide to bid do so using a fixed bidding strategy. Conversely, we identify conditions under which a spot service essentially defaults to an on-demand service, with the provider advertising a single spot price. The findings can help users and providers identify when a spot service is a meaningful addition to cloud service offerings.
"Medial Axes of Voxel Shapes"
Advisor: Tao Ju
The medial axis of a 3d shape is important to understanding the shape's topology and geometry. Currently, the state-of-the-art methods can only compute medial axes with topology guarantee for smooth shapes, i.e. those without sharp edges or corners. Voxel shapes is another important class of shapes widely used in medical imaging. Unfortunately, because they are abundant in sharp edges and corners, there lacks a theoretically sound medial axis algorithm. In this talk, we propose the first algorithm for computing the medial axis of a voxel shape that's both theoretically sound and practically efficient.