在資料中心網路中減少網路流量是一個熱門的議題,對於即時性應用而言,資料聚合是一個經常被使用的方式。資料可以在洗牌階段被聚合, 而且輸出的資料大小可以被減小為輸入的一部分資料大小。儘管如此,現今資料中心的研究並沒有考慮到在時間限制下的資料聚合。在本篇論文中我們運用資料聚合的方法,不只是減少網路流量,同時也減少資料流的延遲使得這些資料流能夠盡量地滿足期限前被收到的需求。我們也利用在以伺服器為主的資料中心的特性並選擇適合的路徑去使得伺服器跟伺服器之間的延遲能夠更低。根據實驗的結果我們發現,這樣以伺服器為主的資料中心聚合樹演算法相較於之前的研究,可以同時提供較低的網路流量並同時維持較低延遲的服務品質。
Bandwidth consumption minimization is a popular issue for data center networks. For real-time applications, data aggregation is commonly used. Data can be aggregated in a shuffle phase, and the output size can be reduced as a fraction of the input size. Nevertheless, existing works on data aggregation do not consider deadline constraints in datacenters. In this paper, we utilize data aggregation to reduce not only bandwidth consumption but also the latency of dataflows so that the deadlines of dataflows can be satifsified. We also exploit the property in server-centric datacenters (SDC) and choose suitable paths to achieve low end-to-end delay. According to the simulation result, the SDC aggregation-tree algorithm can provide both lower bandwidth consumption and lower delay than previous approaches.