在軟體定義網路裡，會有許多需求而傳遞著封包，這些需求所行進的路線可能會遇到一些突發狀況導致路線需要變動，例如：某些節點故障、定期維修檢查或者目前的路線流量已經不太足夠。從一個網路狀態到另一個新的網路狀態我們稱之網路更新。

在傳統上，網路更新採用兩階段更新，所謂的兩階段網路更新是指網路會確保所有的最終規則都被安裝上去後才會把舊的規則從路由器中刪除。這個網路更新的方法是萬能的解，但是卻十分耗費記憶體資源的使用。之後，有人提出了順序規則置換演算法，這個演算法提出若是按照計畫好的順序進行網路更新就可以避免在網路更新時使用過多的記憶體。但是順序規則置換演算法在考慮網路功能的時候會有沒辦法安排出順序的情況，此時網路更新演算法FLIP提供快速的方式判斷部分路由器必須同時安裝兩種規則就可以使用順序規則置換演算法的概念完成網路更新。

網路更新演算法FLIP，在給予網路初始狀態以及最終狀態後會安排網路更新的時程表以及指定需要同時安裝新舊指令的路由器。由此可見FLIP並沒有能力去預防過多的記憶體使用，因此我們這篇論文針對FLIP去重新設計網路最終狀態來使得在進行網路更新的時候避免消耗過多的記憶體資源。

在重新設計網路最終狀態時我們滿足了服務功能鍊的需求。透過與最短滿足服務功能鍊的路徑比較，實驗結果顯示在較小的網路拓樸中，我們最高可以預防52%的記憶體資源使用；在較大的網路拓樸中，我們最高可以預防93%的記憶體資源使用。

In Software Defined Networks, there are many demands that pass packets. The routes of demands may encounter some situations that cause the route to change, such as node failures, periodic maintenance checks, or the capacity of current route traffic not enough. From a network state to another new network state, we call it a network update.

Traditionally, network updates use a two-phase update. The two-phase network update means that the network will ensure that all final rules are installed before the old rules are removed from the router. This method of network update is a general solution, but it is very costly to use memory resources. After two-phase, a ordered rule replacement algorithm was proposed. This algorithm proposes that if the network is updated in the order of planning, it can avoid using too much memory in the network update. However, the order rule replacement algorithm may have no way to arrange the order when considering the network functions. Meanwhile, the network update algorithm FLIP provides a quick way to determine that some routers must install both states rules at the same time to complete the network update by using the concept of ordered rule replacement algorithm.

By giving the network initial state and final state, the network update algorithm FLIP will schedule an order of network updates and will specify some routers that need to install both states rules. This shows that FLIP does not have the ability to prevent excessive memory usage, so this paper redesigned the final state of the network for FLIP to avoid consuming too much memory resources during network updates.

We satisfy the needs of the service function chaining when redesigning the final state of the network. By comparing the shortest path which only consider service chain, the experimental results show that in smaller network topologies, we can prevent up to 52% of usage of memory resources. In larger network topologies, we can prevent up to 93% of usage of memory resources.

[1] H. Vardhan, S. Billenahalli, M. Razo, A. Sivasankar an, L. Tang, P. Montiy, M. Tacca, and A. Fumagaiil. Finding a Simple Path with Multiple Must-include Nodes. In IEEE, 2009.

[2] S. Vissicchio, L. Cittadini. FLIP the(Flow) Table: Fast LIghtweight Policy-preserving SDN Updates. In INFOCOM, 2016.

[3] R. Mahajan and R. Wattenhofer. On Consistent Updates in Software Defined Networks. In HotNets, 2013.

[4] J. Hua, X. Ge, and S. Zhong. FOUM: A Flow-Ordered Consistent Update Mechanism for Software-Defined Networking in Adversarial Settings. In Proc. IEEE INFOCOM, 2016.

[5] G. Sallam, G. Gupta, B. Li, and B. Ji. Shortest Path and Maximum Flow Problems Under Service Function Chaining Constraints. INFOCOM, 2018.

[6] N. Spring, R. Mahajan, and D. Wetherall. Measuring ISP Topologies with Rocketfuel. In SIGCOMM, 2002.

[7] SDN library Website. http://sndlib.zib.de/home.action

[8] VL2: A Scalable and Flexible Data Center Network. Microsoft Research.