Possibilities for parallel computations in utility scale furnace: The code structure analysis
Само за регистроване кориснике
2023
Аутори
Tomanović, IvanBelošević, Srđan
Milićević, Aleksandar
Crnomarković, Nenad
Stojanović, Andrijana
Конференцијски прилог (Објављена верзија)
Метаподаци
Приказ свих података о документуАпстракт
The development of both CPUs and GPUs slowly moved towards the multi-core processors at the start of this century, allowing for true execution of several applications in parallel, or running a single task over multiple threads. This was especially true when considering the affordable and widely available units. The development of powerful GPUs, helped brew an idea in scientific circles to process large and scalable matrices containing mostly non graphical data on the GPUs, the same way they are processing the textures, utilizing the advantage of concurrent execution of many simple operations. Recent advances allowed for even easier and more accessible utilization of new hardware by both the scientific circles and the public in general through the development of new programming tools, languages, and language extensions. In this paper we analyze the structure of an in-house CFD code developed for coal-fired boiler furnace simulations, with an idea to suggest an optimal algorithm structur...e that would allow this code to run on many-core architectures. This is an interesting task, especially given the recent increase in GPU computational abilities, led to the availability of powerful and affordable workstations capable of processing large data sets. Here we analyze the algorithm for a single thread optimized code, identifying potential problems and recursive dependencies, and suggest the structure that would allow movement towards the parallel versions of existing code and its execution on both multi- and many-core architectures. Goal of this analysis is to identify and understand the inherently recursive portions of the code that should undergo modifications to allow for fully parallel executions in future efficient simulations of boiler furnaces.
Кључне речи:
OpenACC / GPU / CFD / parallel algorithms / utility boiler furnaceИзвор:
International Conference Power Plants 2023 : Proceedings, 2023, 763-771Издавач:
- Belgrade : Society of Thermal Engineers of Serbia
Напомена:
- Power Plants 2023 : Elektrane 2023; November 8-10, 2023, Zlatibor, Serbia
Колекције
Институција/група
VinčaTY - CONF AU - Tomanović, Ivan AU - Belošević, Srđan AU - Milićević, Aleksandar AU - Crnomarković, Nenad AU - Stojanović, Andrijana PY - 2023 UR - https://vinar.vin.bg.ac.rs/handle/123456789/12768 AB - The development of both CPUs and GPUs slowly moved towards the multi-core processors at the start of this century, allowing for true execution of several applications in parallel, or running a single task over multiple threads. This was especially true when considering the affordable and widely available units. The development of powerful GPUs, helped brew an idea in scientific circles to process large and scalable matrices containing mostly non graphical data on the GPUs, the same way they are processing the textures, utilizing the advantage of concurrent execution of many simple operations. Recent advances allowed for even easier and more accessible utilization of new hardware by both the scientific circles and the public in general through the development of new programming tools, languages, and language extensions. In this paper we analyze the structure of an in-house CFD code developed for coal-fired boiler furnace simulations, with an idea to suggest an optimal algorithm structure that would allow this code to run on many-core architectures. This is an interesting task, especially given the recent increase in GPU computational abilities, led to the availability of powerful and affordable workstations capable of processing large data sets. Here we analyze the algorithm for a single thread optimized code, identifying potential problems and recursive dependencies, and suggest the structure that would allow movement towards the parallel versions of existing code and its execution on both multi- and many-core architectures. Goal of this analysis is to identify and understand the inherently recursive portions of the code that should undergo modifications to allow for fully parallel executions in future efficient simulations of boiler furnaces. PB - Belgrade : Society of Thermal Engineers of Serbia C3 - International Conference Power Plants 2023 : Proceedings T1 - Possibilities for parallel computations in utility scale furnace: The code structure analysis SP - 763 EP - 771 UR - https://hdl.handle.net/21.15107/rcub_vinar_12768 ER -
@conference{ author = "Tomanović, Ivan and Belošević, Srđan and Milićević, Aleksandar and Crnomarković, Nenad and Stojanović, Andrijana", year = "2023", abstract = "The development of both CPUs and GPUs slowly moved towards the multi-core processors at the start of this century, allowing for true execution of several applications in parallel, or running a single task over multiple threads. This was especially true when considering the affordable and widely available units. The development of powerful GPUs, helped brew an idea in scientific circles to process large and scalable matrices containing mostly non graphical data on the GPUs, the same way they are processing the textures, utilizing the advantage of concurrent execution of many simple operations. Recent advances allowed for even easier and more accessible utilization of new hardware by both the scientific circles and the public in general through the development of new programming tools, languages, and language extensions. In this paper we analyze the structure of an in-house CFD code developed for coal-fired boiler furnace simulations, with an idea to suggest an optimal algorithm structure that would allow this code to run on many-core architectures. This is an interesting task, especially given the recent increase in GPU computational abilities, led to the availability of powerful and affordable workstations capable of processing large data sets. Here we analyze the algorithm for a single thread optimized code, identifying potential problems and recursive dependencies, and suggest the structure that would allow movement towards the parallel versions of existing code and its execution on both multi- and many-core architectures. Goal of this analysis is to identify and understand the inherently recursive portions of the code that should undergo modifications to allow for fully parallel executions in future efficient simulations of boiler furnaces.", publisher = "Belgrade : Society of Thermal Engineers of Serbia", journal = "International Conference Power Plants 2023 : Proceedings", title = "Possibilities for parallel computations in utility scale furnace: The code structure analysis", pages = "763-771", url = "https://hdl.handle.net/21.15107/rcub_vinar_12768" }
Tomanović, I., Belošević, S., Milićević, A., Crnomarković, N.,& Stojanović, A.. (2023). Possibilities for parallel computations in utility scale furnace: The code structure analysis. in International Conference Power Plants 2023 : Proceedings Belgrade : Society of Thermal Engineers of Serbia., 763-771. https://hdl.handle.net/21.15107/rcub_vinar_12768
Tomanović I, Belošević S, Milićević A, Crnomarković N, Stojanović A. Possibilities for parallel computations in utility scale furnace: The code structure analysis. in International Conference Power Plants 2023 : Proceedings. 2023;:763-771. https://hdl.handle.net/21.15107/rcub_vinar_12768 .
Tomanović, Ivan, Belošević, Srđan, Milićević, Aleksandar, Crnomarković, Nenad, Stojanović, Andrijana, "Possibilities for parallel computations in utility scale furnace: The code structure analysis" in International Conference Power Plants 2023 : Proceedings (2023):763-771, https://hdl.handle.net/21.15107/rcub_vinar_12768 .