軟體定義網路與網路功能虛擬化下整合資料、控制與管理層之階層式資料中心架構

Abstract

傳統網路上，網路供應商難以管理虛擬化網路服務(NFV)於自己的網路環境，與供應給用戶選擇所需求的網路服務。我們設計了階層式NFV資料中心架構整合軟體定義網路與NFV，整體連結成多棵樹狀的拓樸。在此架構上可以探討兩個重要的議題: 租戶所需求的服務應在哪一層的資料中心提供與階層式架構下資源的擺放。前者我們提出最小延遲的演算法並且使用全域的伺服器週期性的收集當下各資料中心中服務串接的處理時間，挑出最適合的資料中心位置。後者我們提出四種可能的分配策略探討在我們的架構下不同的資源擺放的影響。在實驗上我們比較了傳統與我們架構的差異與是否適合NFV服務的擺放。在我們的架構下，三層的階層式架構有最小的NFV服務的回應時間在伺服器高負荷的情況下，並且在於較短的路徑與良好的資源擺放可以有效地降低NFV服務的回應時間。流量的來源與目的地將會明顯的影響選擇的方針，決定是否需要考慮額外的路徑所造成的網路延遲，我們將它分為三個情況: 在S&S的情況下，最好的資源分配方案為heavy feet，有67%的需求導向第一層執行NFV的服務;在S&D的情況下最好的方案Avg. dc在於第一層與第二層分別有32%與56%的使用率;D&D的情況下最好的方案是Avg. tier，此方案將會讓各層的使用量平均。

Traditionally, network operator is hard to launch new network functions and care about NFV service allocation. Therefore, we design the single NFV datacenter with integrated SDN-NFV management and then design a hierarchical NFV datacenter architecture in which datacenters are connected into a multi-tree topology. In a hierarchical-datacenter architecture, there are two important issues: tenant’s services should be served in which datacenter and how many service capacities are allocated in each tier. For the former, we propose an approach, called minimum-delay selection, that selects the datacenter which performs the services with the minimum latency. For the latter, we propose four possible allocation policies depending on different scenarios. Our experiment results show that a 3-tier hierarchical-datacenter architecture produces a low NFV service response time and high server utilization, because it has short propagation delay and workload can be balanced among datacenters in different tiers. The source and the destination of the subscriber network’s traffic will significantly affect the suitable allocation policy, deciding whether or not to weight the propagation factor more heavily in the datacenter selection decision, we classify three cases: In S&S case, heavy feet is the best method of resource allocation with 67% NFV services performed by tier 1, the D&S case is the method of Avg. dc is better choice with 32% in tier 1 and 56% in tier 2; D&D uses the method of Avg. tier is uniformly distributing to all tiers.