A node selection algorithm for hybrid peer-to-peer services based on a modified dueling-DQN model

• Item type: SIG Technical Reports(1)
• 公開日: 2023-06-27
• タイトル: A node selection algorithm for hybrid peer-to-peer services based on a modified dueling-DQN model
• タイトル / 言語: en
• タイトル: A node selection algorithm for hybrid peer-to-peer services based on a modified dueling-DQN model
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: ネットワーク・トラフィック
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: NARA Institute of Science and Technology
• 著者所属: NARA Institute of Science and Technology
• 著者所属: NARA Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): NARA Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): NARA Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): NARA Institute of Science and Technology
• 著者名: Shan, Gao; Kohei, Ichikawa; Hajimu, Iida
• 著者名(英): Shan, Gao; Kohei, Ichikawa; Hajimu, Iida
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Hybrid Peer to Peer (P2P) file sharing combines centralized and decentralized architectures, where a central server maintains the node list while data is transferred in a P2P manner. The hybrid P2P structure allows each node to decide from which node to download data, but this can result in slower and inefficient data exchange. To address this issue, some P2P file sharing networks like BitTorrent use the Tit-for-Tat (TFT) algorithm, which selects nodes based on their past interactions. However, TFT may not work effectively under rapidly changing network conditions. In this paper, we propose a modified node selection algorithm based on the dueling DQN model, which considers real-time states. The central server maintains node addresses and collects real-time status information such as CPU, memory utilization, and network latency. The modified dueling DQN model returns the node with the highest overall rating, and data is downloaded from the selected node. The results show that our proposed model is able to achieve almost the same download time as the TFT algorithm, while our model can adapt to the network environment faster than the TFT algorithm when the network latency fluctuates significantly.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Hybrid Peer to Peer (P2P) file sharing combines centralized and decentralized architectures, where a central server maintains the node list while data is transferred in a P2P manner. The hybrid P2P structure allows each node to decide from which node to download data, but this can result in slower and inefficient data exchange. To address this issue, some P2P file sharing networks like BitTorrent use the Tit-for-Tat (TFT) algorithm, which selects nodes based on their past interactions. However, TFT may not work effectively under rapidly changing network conditions. In this paper, we propose a modified node selection algorithm based on the dueling DQN model, which considers real-time states. The central server maintains node addresses and collects real-time status information such as CPU, memory utilization, and network latency. The modified dueling DQN model returns the node with the highest overall rating, and data is downloaded from the selected node. The results show that our proposed model is able to achieve almost the same download time as the TFT algorithm, while our model can adapt to the network environment faster than the TFT algorithm when the network latency fluctuates significantly.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA12326962
• 書誌情報: 研究報告インターネットと運用技術（IOT）; 巻 2023-IOT-62, 号 9, p. 1-6, 発行日 2023-06-27
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8787
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会