A Hybrid Computing Infrastructure for Climate Simulation
| dc.contributor.advisor | Yang, Chaowei | |
| dc.contributor.author | Liu, Kai | |
| dc.creator | Liu, Kai | |
| dc.date.accessioned | 2018-10-22T01:19:51Z | |
| dc.date.available | 2018-10-22T01:19:51Z | |
| dc.date.issued | 2017 | |
| dc.description.abstract | The climatological community relies increasingly on computing intensive models and applications to study atmospheric chemistry, aerosols, carbon cycle and other tracer gases. These models and applications are becoming increasingly complex and bring computing challenges including: 1) the enormous computational power required for running these models and applications to produce results in a reasonable timeframe; 2) the challenging in providing convenient and fast solutions distributing and storing the massive climate model outputs; 3) the lack of methods for visualizing the climate simulation results efficiently and reliably. Volunteer computing provides a potential solution for tackling the computational power problem by obtaining large amounts of computational resources from global volunteers. Meanwhile, virtualization technology allows researchers to run climate models in a predefined virtual machine. Cloud computing storage provides advantages for distributing and storing climate data and outputs with low-cost. The Load Balancer and Auto Scaling in cloud computing provides a good solution to visualize the climate simulation results. This dissertation reports our research on integrating and optimizing volunteer computing, virtualization technology and cloud computing for climate simulation by: 1) using volunteer computing resources to leverage large number of home computers to support climate simulations; using virtualization technology to enable the climate models run on heterogeneous computers while providing bit-level homogeneous computing environment; optimizing the output collection mechanism to periodically upload climate model output; and optimizing the credit system to grant credits periodically to volunteers for volunteer retention to support long time climate simulation tasks. 2) Using cloud Simple Storage Service provided by leading cloud providers to develop a global replication storage to distribute cloud models and data to global volunteers. 3) Using Load Balancing to distribute incoming WMS requests across multiple cloud instances to improve the performance; Using Auto Scaling to help to maintain climate visualization availability and allows climate scientists to dynamically scale the cloud capacity. A prototype is developed to demonstrate the feasibility and efficiency of proposed techniques. The prototype is further tested in the Climate@Home project, a hybrid computing project using volunteer computing and cloud computing. Result shows that this research provides a computationally efficient and usable approach to accelerate climate simulation. | |
| dc.format.extent | 83 pages | |
| dc.identifier.uri | https://hdl.handle.net/1920/11266 | |
| dc.language.iso | en | |
| dc.rights | Copyright 2017 Kai Liu | |
| dc.subject | Geographic information science and geodesy | |
| dc.subject | Climate Simulation | |
| dc.subject | Cloud Computing | |
| dc.subject | Hybrid Computing | |
| dc.subject | Volunteer Computing | |
| dc.title | A Hybrid Computing Infrastructure for Climate Simulation | |
| dc.type | Dissertation | |
| thesis.degree.discipline | Earth Systems and Geoinformation Sciences | |
| thesis.degree.grantor | George Mason University | |
| thesis.degree.level | Ph.D. |
Files
Original bundle
1 - 1 of 1