Distributed Low Latency Computing With OpenCL: A Scalable Multi-Access Edge Computing Framework
Solanti, Jan (2020)
Solanti, Jan
2020
Tietotekniikan DI-ohjelma - Master's Programme in Information Technology
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2021-01-08
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202012088592
https://urn.fi/URN:NBN:fi:tuni-202012088592
Tiivistelmä
The ever increasing computational complexity of applications requires increasing amounts of processing power, yet users are increasingly moving to resource and power constrained mobile devices for their computational needs. This calls for creative solutions that provide increased processing capabilities without impacting battery life or degrading the user experience. Multi-Access Edge Computing is a standardization effort to provide consistent cloud edge environments for optimizing applications on low-power devices by enabling developers to offload parts of the application to networked computing infrastructure that is located physically close to the device running the application.
This master’s thesis describes pocl-r, a framework for transparently offloading computation in applications that use the OpenCL API for heterogeneous computation. The implementation performs comparably to previous work in synthetic benchmarks while offering greater flexibility to application developers by not depending on 3rd party communication frameworks and not requiring the application to be aware of any particular OpenCL API extensions. In addition to synthetic benchmarks, the impact of offloading heavy computation is measured in a case study of a mobile application that renders a streamed animated point cloud. The resulting energy consumption when offloading was measured to be roughly half of what it was without offloading. When additionally making the application aware of a minimal extension to the OpenCL API, energy consumption per frame was cut to a roughly a 20th of the original while also increasing the framerate tenfold.
This master’s thesis describes pocl-r, a framework for transparently offloading computation in applications that use the OpenCL API for heterogeneous computation. The implementation performs comparably to previous work in synthetic benchmarks while offering greater flexibility to application developers by not depending on 3rd party communication frameworks and not requiring the application to be aware of any particular OpenCL API extensions. In addition to synthetic benchmarks, the impact of offloading heavy computation is measured in a case study of a mobile application that renders a streamed animated point cloud. The resulting energy consumption when offloading was measured to be roughly half of what it was without offloading. When additionally making the application aware of a minimal extension to the OpenCL API, energy consumption per frame was cut to a roughly a 20th of the original while also increasing the framerate tenfold.