报告人: Prof. Lei Jiao, University of Oregon, USA
联系人: 陈阳 firstname.lastname@example.org
The cloud infrastructure is transforming into a coexistent architecture of the conventional gigantic clouds at the Internet core plus the highly distributed, small-scale clouds at the network edge in close proximity to the end users. To leverage the great potential of this emerging platform, the service provider confronts the problem of the dynamic allocation of cloud resources, which is particularly challenging due to multiple factors: the reconfiguration cost that couples consecutive time slots, the dynamic, unpredictable service demands, the heterogeneity of the distributed resources, and the arbitrary mobility of the end users. Considering user demands from diverse regions, we design a novel online algorithm that decouples the problem over time by constructing a series of related subproblems solvable at each corresponding time slot using the output of the previous time slot. Via solid formal analysis, we prove that, without prior knowledge beyond the current time slot, our online algorithm can provide a solution on the fly with a parameterized competitive ratio for arbitrary workload and resource price dynamics. Further, considering user mobility and workload migration across the edge clouds, we derive a gap-preserving transformation of the problem, and exhibit that our online algorithmic framework and the theoretical analysis still apply, preserving a similar parameterized competitive ratio while only introducing an additional constant. Our evaluations driven by real-world data traces demonstrate that our online algorithm achieves up to 9× total cost reduction compared to the sequence of greedy, one-shot optimizations and at most 3× the offline optimum. With user mobility, it achieves near-optimal results, reducing the total cost by up to 4× compared to static approaches and outperforming one-shot optimizations by up to 70%.
Lei Jiao is an assistant professor at the Department of Computer and Information Science, University of Oregon. He was a member of technical staff at Bell Labs in Dublin, Ireland from 2014 to 2016. He received the Ph.D. degree in computer science from University of Göttingen, Germany in 2014. He worked as a researcher at IBM Research in Beijing, China in 2010. His research interests span broadly optimization and control of systems and networks, with a recent focus on the problem domains of edge computing, cloud data center computing, and online social networking. His research has been published in IEEE/ACM Transactions on Networking, IEEE INFOCOM, ICNP, IPDPS, ICDCS, etc. He is the recipient of the Best Paper Award of IEEE LANMAN 2013.