TY - GEN
T1 - P2PRC—A Peer-To-Peer Network Designed for Computation
AU - Selvacoumar, Akilan
AU - Soobhany, Ahmad Ryad
AU - Reji, Benjamin Jacob
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.
PY - 2022/4/21
Y1 - 2022/4/21
N2 - This paper focuses on developing the peer-to-peer rendering and computation (P2PRC) framework, which is a distributed framework for executing computationally demanding tasks that a personal machine with limited processing power will struggle to run such as graphically demanding video games, rendering 3D animations, and protein folding simulations. A custom peer-to-peer network was implemented to decentralize the execution of tasks either on central processing unit (CPU) or graphical processing unit (GPU), in order to increase the bandwidth for running tasks. To prevent the tasks in the peer-to-peer network from corrupting the operating system (OS) of the server, they will be executed in a virtual environment in the server. The user acting as the client is provided full flexibility on how to batch the tasks, and the user acting as the server has complete flexibility on tracking the container’s usage and killing the containers at any time. The effectiveness of the network and the performance of the distributed task execution of the distributed framework were evaluated using Horovod and TensorFlow benchmarks. Preliminary results are very promising with 86 and 97% improvements for CPU and GPU distribution, respectively.
AB - This paper focuses on developing the peer-to-peer rendering and computation (P2PRC) framework, which is a distributed framework for executing computationally demanding tasks that a personal machine with limited processing power will struggle to run such as graphically demanding video games, rendering 3D animations, and protein folding simulations. A custom peer-to-peer network was implemented to decentralize the execution of tasks either on central processing unit (CPU) or graphical processing unit (GPU), in order to increase the bandwidth for running tasks. To prevent the tasks in the peer-to-peer network from corrupting the operating system (OS) of the server, they will be executed in a virtual environment in the server. The user acting as the client is provided full flexibility on how to batch the tasks, and the user acting as the server has complete flexibility on tracking the container’s usage and killing the containers at any time. The effectiveness of the network and the performance of the distributed task execution of the distributed framework were evaluated using Horovod and TensorFlow benchmarks. Preliminary results are very promising with 86 and 97% improvements for CPU and GPU distribution, respectively.
KW - Distributed computation
KW - P2P networks
KW - P2PRC
KW - Virtualization
UR - http://www.scopus.com/inward/record.url?scp=85129266069&partnerID=8YFLogxK
U2 - 10.1007/978-981-16-7618-5_45
DO - 10.1007/978-981-16-7618-5_45
M3 - Conference contribution
AN - SCOPUS:85129266069
SN - 9789811676178
T3 - Lecture Notes in Networks and Systems
SP - 517
EP - 526
BT - Proceedings of International Conference on Information Technology and Applications. ICITA 2021
A2 - Ullah, Abrar
A2 - Gill, Steve
A2 - Rocha, Álvaro
A2 - Anwar, Sajid
PB - Springer
T2 - 15th International Conference on Information Technology and Applications 2021
Y2 - 13 November 2021 through 14 November 2021
ER -