Loading...

Proposing an Execution Architecture to Reduce the Cost of Stream Processing in FPGA-Enabled Cloud

Nasiri, Hamid | 2023

0 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 58159 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Goudarzi, Maziar
  7. Abstract:
  8. Today, with the expansion of the Internet of Things, use of various sensors and need for real-time processing of streams of data is felt more than ever. This need has forced organizations to continuously upgrade processing hardware. The high cost of providing equipment, the difficult management of heterogeneous processing resources and the variable processing power required at different times are the most important factors for companies to turn to cloud platforms. One of the main challenges of using cloud resources is the waste of a significant part of the processing power of the rented resources. The most important factors of this wastage are the inadequacy of the reserved resources with the required processing power over time and the lack of proper productivity of the resources. From the point of view of the data center, problems such as interference between virtual machines and the failure of physical machines prevent the accurate determination of the processing power provided by virtual machines; which aggravates the first factor. In addition, in stream processing the input rate can change over time, which makes it more difficult to determine the amount of required processing power. In this thesis, focusing on stream processing in cloud environments, it is tried to minimize the final cost of processing by properly estimating the programs' required processing power in real-time and using heterogeneous processing resources according to the type of tasks. The conscious use of the unique characteristics of cloud resources, resource heterogeneity and re-configurable chips has increased the efficiency of the proposed method compared to similar methods. Based on the results of the experiments, in a heterogeneous environment, the conscious allocation of resources to program tasks, while increasing resource efficiency, leads to a reduction of 8-46% in processing delay compared to the best competing algorithm and reduces the processing cost of streaming applications from 10-42%. As a result, the presented architecture can reduce the processing cost of stream applications by its efficient elastic resource allocation mechanism
  9. Keywords:
  10. Cloud Processing Platforms ; Resources Allocation ; Reinforcement Learning ; Data Stream Processing ; Reconfigurable Devices

 Digital Object List

 Bookmark

...see more