隨著圖神經網絡（GNNs）的興起與其強大的解釋能力，對於高效存取圖 結構資料的需求重新變的重要。然而圖結構資料不規則且複雜的特性加劇了 寫入放大和讀取的延遲。儘管如此，現階段的針對圖儲存相關的研究主要集 中在靜態圖上。雖然一些研究涉及動態圖，它們討論的是在不同的時間戳下 保持該時間點下的圖，而非真正的去維護一個會隨時間而變動的單一圖資料 結構。 因此，本項研究提出了一個關注單一圖結構資料隨時間演變的局部感知 圖快閃記憶體－LaGRange的全面設計。它能夠更有效地管理圖的更新，同時 減輕寫入放大和讀取延遲。;With the rise of Graph Neural Networks (GNNs), the need for efficient graph storage has resurfaced. While retrieving data from external storage, the irregular pattern of graph data exacerbates both write amplification and read latency. Despite this, current research on graphs predominantly focuses on static graphs. Although some studies address dynamic graphs, it maintains multiple versions of a graph at different timestamps rather than a single graph evolving over time. The study proposes Locality-aware Graph SSD with Range(LaGRange) maintenance which focuses on a single graph that evolves over time. It’s a comprehensive design to manage graph updates effectively, mitigating write amplification and read latency simultaneously.
