Non-congestion datacenter networks based on the distributed rebalancing algorithm

2021 Fiscal Year Final Research Report

• Project/Area Number: 19K11928
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 60060:Information network-related
• Research Institution: Toyama Prefectural University
• Principal Investigator: Ohta Satoru 富山県立大学, 工学部, 教授 (80438168)
• Project Period (FY): 2019-04-01 – 2022-03-31
• Keywords: データセンタ / スイッチ網 / ネットワーク / ルーティング / スループット

Outline of Final Research Achievements

For the purpose of establishing non-congestion data center networks with low control overhead, this study investigated the distributed rebalancing algorithm, which evenly diffuses flows in a Clos network with employing locally obtainable information. The study first clarified the throughput improvement by the algorithm through packet-level computer simulation. Then, the algorithm was implemented by software. This implementation runs on Linux PCs, which emulate switches of a Clos network. The implementation confirms the feasibility of the algorithm. The implemented algorithm was tested on an experimental Clos network, and compared with an existing technique: random routing. The result shows that the rebalancing algorithm outperforms the random routing technique. Meanwhile, fundamental investigations were also carried out on Clos networks. These investigations include the problem to rearrangements needed for unblocking, and efficient method to control elastic optical switching networks.

Free Research Field

情報ネットワーク

Academic Significance and Societal Importance of the Research Achievements

本研究の成果，特に実装を行ったことで，Clos網に基づく現実のデータセンタ・ネットワークに対する，分散リバランシング制御が適用可能性について見通しがついた．また，従来技術よりも優れたスループット特性を達成可能なことを計算機シミュレーションと実験で確認したので，分散リバランシング制御はClos網の伝送能力を最大限引き出すことが可能であることを示した．これは，多くのITサービスがデータセンタを介して提供される現状では，安定した品質良いサービスの提供上，有意義である．特にビッグデータを利用するアプリケーションで要求される大きな伝送帯域の達成のため有益な結果である．