在具有流量彙整能力的波長分多工網路下，許多低頻寬的需求連線可以彙整在同一條高容量的波長上同時傳輸，然後再各自拆解至傳輸目的地，以提高網路頻寬的使用率，但是這些需求連線會因為要求頻寬的多寡不同而使得其被阻塞的機率有不公平的情況產生，在此本篇論文定義差異性服務意即系統可提供給需求連線不同要求頻寬的服務，於是要如何使得系統提供一個公平的差異性服務便是本篇論文著重的目標。
本篇論文使用之前研究所定義的馬可夫決策過程來描述在單一連結且單一波長下具有流量彙整能力的波長分波多工網路如何對需求連線實施公平允入控制，並在系統加入了一個只能儲存單一需求連線的暫存器將其列入馬可夫決策過程的算式中，若該暫存器內存有需求連線則系統就無法再接受後來要求服務的需求連線直到暫存器內的需求連線被服務，設置此暫存器的目的是為了增加系統的公平性以及提高網路產量。上述馬可夫決策過程的目標是為了最大化單一連結且單一波長的網路下系統頻寬的使用率，我們使用循環策略演算法求出符合該系統的最佳策略，接著本篇論文提出了一個以該最佳策略為基準的公平允入控制演算法將單一連結且單一波長網路的公平性延伸到多重連結且多重波長的網路上。本篇演算法的主要概念在於將需求連線的連線路徑視為多個單一連結且單一波長之子系統的集合，在每個子系統內我們預設以馬可夫決策過程所求出的最佳策略當作公平允入控制的指標，但為顧及整體網路的公平性我們也有可能會去調整某些子系統的最佳策略，使得系統達到全域性的公平。
最後我們會以一個6×6的網柱狀網路拓撲來模擬網路實際的運作情形。我們將分別針對本篇方法與單純使用馬可夫決策過程求出最佳策略、數學統計方法、以及使用暫存器等這些公平允入控制的方法，來比較效能上的差別。在本篇論文中，我們定義公平率為不同頻寬需求的需求連線裡，其被阻塞的機率中取最大值和最小值之比值，來當作判斷整體網路是否公平的依據。實驗證明本篇論文提出的方法並不會使得整體需求連線被阻塞的機率大幅上升，而且系統使用本篇方法來實作公平允入控制，其公平率會比單純使用馬可夫決策過程的方法改善了約53%，比數學統計的方法最多改善了約77%，而比使用暫存器中的方法最多改善了約44%。此外我們並觀察系統是否對所有類別連線作公平允入控制時其類別連線被阻塞機率的變化，我們還比較使用本篇方法與其他方法其暫存器內的需求連線被搶先的個數，並且將本篇演算法搭配不同的波長配置法來觀察公平率的變化，最後比較本篇方法在不同的網路拓樸下其公平率的差別，我們將以這些數據結果來驗證我們所提出方法的優勢所在。

In traffic groomed optical networks, many requests with low bandwidth requirement can be multiplexed/demultiplexed into/from a wavelength with large capacity, in order to improve bandwidth utilization. Requests of higher bandwidth requirement may encounter higher blocking probabilities which results in capacity unfairness. In this paper, a Markov Decision Process (MDP) is used to provide a fairness control for differentiated services in traffic groomed optical networks. The differentiated services in this paper are defined to be the services with different bandwidth requirements.
Based on MDP formulation which maximizes the overall utilization for different traffic classes in a single-link single-wavelength network, a Call Admission Control (CAC) can be determined by the Policy Iteration method as a function of current capacity usage on each wavelength. Besides, a buffer is employed in order to hold a request rather than directly reject it when the residual bandwidth is insufficient in the network. Since the routing path of a request is composed of several single-hops, no request can be admitted unless each single-hop of a request is admitted. We propose a policy-based CAC algorithm which is base on the CAC policy derived from MDP as the CAC control on each single-hop and extend it to the multiple-hop multiple-wavelength networks. When all single-hops on the routing path can afford the request bandwidth, the request will be accepted if it is treated unfairly or it has passed through CAC control. Otherwise the request will be put into the buffer of those single-hops on its routing path or rejected due to lack of available bandwidth. The request will consume the bandwidth if it is accepted on each single-hop on its routing path on the same wavelength to obey the wavelength continuity constraint.
We first use a 6×6 mesh-torus network topology to compare the proposed policy-based CAC algorithm with other CAC mechanisms. To obtain a fairness index, we defined the fairness ratio as the ratio of the largest class-based blocking probability to the smallest one because the fairness is achieved when the requests of all classes have the same blocking probability under the assumption that all requests of different classes demand the same amount of bandwidth. Simulation results show that the proposed algorithm improves up to 77% in the fairness ratio compared with other fair admission control methods. Furthermore, the proposed algorithm will not dramatically increase the overall blocking probability of the network. Additionally, the proposed algorithm is the best among all CAC methods in terms of the network throughput. Finally, we use different wavelength assignment algorithms and different network topologies to analyze the impact on the performance in terms of fairness ratio and blocking probability.

[1] B. Mukherjee, et al. “Traffic grooming in mesh optical networks,” OFC-2004, Vol. 2, pp. 23-27, Mar. 2004.
[2] K. Zhu, H. Zang, B. Mukherjee, “A comprehensive study of next-generation optical grooming switches,” IEEE JSAC, Vol. 21, no. 7, pp. 1173-1186, Sep. 2003.
[3] A. K. Somani, “Survivability and Traffic Grooming in WDM Optical Networks,” Cambridge University Press 2006.
[4] K. Zhu, B. Mukherjee, ”A Review of Traffic Grooming in WDM optical networks: Architectures and Challenges,” Optical Networks Magazine, vol. 4, no. 2, pp. 55-64 , Mar./Apr. 2003.
[5] S. Thiagarajan , A. K. Somani, ”Capacity Fairness of WDM Networks with grooming capabilities,” Optical Network Magazine, Vol. 2, no. 3, pp. 24-32, May/June 2001.
[6] L. Yuan-Cheng, L. Yu-Dar, “A Fair Admission Control for Large-Bandwidth Multimedia Applications,” ICDCSW, pp. 317-322 , 2002.
[7] C. S. R. Murthy, M. Gurusamy, “WDM Optical Networks: Concepts, Design, and Algorithms,” PHPTR Press, Nov. 2001.
[8] M. Sivakumar, K.M. Sivalingam, S. Subramaniam, “On factors affecting the performance of dynamically groomed optical WDM mesh networks ,“ HPSR. 2005, Vol.1, pp. 411- 415 May 2005.
[9] K. Zhu; H. Zhu; B. Mukherjee, “Traffic engineering in multigranularity heterogeneous optical WDM mesh networks through dynamic traffic grooming,” MNET , 2003, Vol. 17, no 2, pp.8-15, Mar./ Apr. 2003 .
[10] K.W. ,Ross , D.H.K., Tsang , “The Stochastic Knapsack Problem,” IEEE Transaction on Communications, Vol. 37, no.7, pp. 740-747, July 1989.
[11] I. Cerutti, P. Sudhakar, A. Fumagalli, “A Threshold-Based Blocking Differentiation Mechanism for Networks with Wavelength Continuity Constraint,” Transparent Optical Networks, 2005, Vol.1, pp. 179- 182, July 2005.
[12] J. Choi, T. Kwon, Y. Choi, M. Naghshineh, “Call admission control for multimedia service in mobile cellular networks: a Markov decision approach,” IEEE ISCC 2000, Antibes, pp. 594-599, July 2000.
[13] R, A, Howard, “Dynamic Programming and Markov Process”, M.I.T. Press, Cambridge,1960
[14] K. Mosharaf, J. Talim, L. Lambadaris, L. Marmorkos, "Service differentiation and fairness control in WDM grooming networks", GLOBECOM ’04, Vol. 3, pp.1968–1973, Nov. 2004.
[15] D. P. Bertsekas, ”Dynamic programming deterministic and stochastic models,” PHPTR Press, , Apr. 1987.
[16] S. Subramaniam, R.A. Barry, “Wavelength Assignment in Fixed Routing WDM Networks,” ICC 97, Vol.1, pp. 406-410 , 1997.