Joint bandwidth allocation on dedicated and shared wavelengths for QoS support in multi-wavelength optical access network

Transcription

1 Published in IET Communications Received on 4th March 2013 Revised on 12th May 2013 Accepted on 31st May 2013 ISSN Joint bandwidth allocation on dedicated and shared wavelengths for QoS support in multi-wavelength optical access network Cuiping Ni, Chaoqin Gan, Haibin Chen Key Laboratory of Specialty Fiber Optics and Optical Access Networks, Shanghai University, Shanghai, People s Republic of China cqgan@shu.edu.cn Abstract: Wavelength-division multiplexing (WDM) technology has been recognised as one of the most promising solutions in optical access network. A general ring-based multi-wavelength optical access network based on WDM technology, which can realise triple-play service including point-to-point service and multicast service is introduced in this study. To effectively utilise the network bandwidth resources, the authors propose a joint bandwidth allocation scheme on dedicated and shared wavelengths. The proposed scheme focuses on quality of services (QoSs) support. It is carried out according to the priority queue discipline to support QoS in terms of average packet delay, packet delay variance and throughput. Moreover, with the fair-awareness in the scheduling scheme, the lower priority services can be prevented from starvation and the over-allocation problem can be avoided. The simulation results show that the proposed scheme can reduce average packet delay and packet delay variance for high-priority services to ensure QoS. 1 Introduction With the tremendous growth of bandwidth demand and the emergence of new bandwidth intensive applications, wavelength-division multiplexing (WDM) technology has been regarded as an ideal solution in optical access network [1, 2]. For the next generation access network, huge bandwidth, high reliability and metro-access are of great importance. Hence, ring-based topologies with multiple wavelengths are widely researched and considered as the potential architecture for next generation WDM architecture [3 5]. Besides, considering the triple-play service (TPS) that delivers integrated voice, data and video services, multicast transmission has become a very important requirement for access networks. Some solutions suggest amplitude shift keying/differential phase shift keying (ASK/DPSK) orthogonal modulation to realise multicast function [6, 7]. However, they are very expensive because a DPSK receiver must be configured in each optical network unit (ONU) in orthogonal modulation schemes. In [8], an arrayed waveguide grating (AWG)-based WDM passive optical network (PON) architecture with multicast capability was proposed. The two-stage AWGs provided both high scalability and transparency for each multicast group. A WDM PON architecture that reserves a single downstream wavelength [9] has been proposed to broadcast downstream signals to all ONUs. Among all the architectures mentioned above, the transmission mechanism and bandwidth allocation need further study. Thus, a general ring-based multi-wavelength optical access network based on WDM technology, which can realise TPS including point-to-point service and multicast service is introduced here for consistency. Compared with the other architectures mentioned above, the general network architecture in our paper is higher cost-effective, more scalable, and easier to upgrade. What s more, this architecture can realise effective dynamic bandwidth allocation. In access network, the dynamic bandwidth allocation (DBA) plays a crucial role in efficiently and fairly allocating the bandwidth among all users [10 12]. Moreover, the quality of service (QoS) is also an essential requirement to support the emerging applications such as voice communications (VoIP), video conferencing (interactive video) and data traffic. To date, various DBA algorithms to provide differentiated classes of service for access network have been proposed. A bandwidth guaranteed polling is proposed in [13] for supporting QoS. QoS-based queue size prediction mechanisms have been proposed for differentiated services [14, 15]. Schemes that can guarantee jitter [16] and fairness [17] for QoS support have been proposed. Hwang et al. proposed a generic QoS-aware interleaved DBA to eliminate the idle period, enhance QoS and effectively reduce high-priority service delay and jitter [18]. Furthermore, many bandwidth scheduling scheme are proposed based on specific network architecture. A distributed ethernet passive optical network (EPON) architecture that supports differentiated services through the integration of inter-onu scheduling and intra-onu scheduling is proposed in [19]. Kiaei et al. proposed a scheduling and bandwidth allocation approach on the co-existence of 10G-EPON and WDM PONs [20] 1863

2 and Hara et al. proposed a 100 Gbit/s-class-WDM/ TDM-PON system that uses wavelength tunability and DBA to provide a flexible load balance and enlarge total bandwidth [21]. A novel dynamic wavelength scheduled hybrid WDM/TDM PON is proposed for aggregating large files in distributed computing applications in [22]. In this paper, we introduce a general multi-wavelength optical access network architecture realising triple-play services including point-to-point service and multicast service. The general network architecture features large capacity, high security and easy upgrade. Thus, it has a wide range of application. Furthermore, we investigate the transmission mechanism and bandwidth allocation in such a general multi-wavelength optical access network. A joint bandwidth allocation scheme on dedicated and shared wavelength is proposed to effectively utilise the network bandwidth resources. Unlike other architecture-based bandwidth scheduling scheme mentioned before [19 22], our scheme has a wider range of application. Moreover, the proposed scheme focuses on QoSs support because of the increasing importance of QoS in DBA as discussed before [13 18]. It is carried out according to the priority queue discipline to support QoS in terms of average packet delay, packet delay variance and throughput. Additionally, the threshold and ratio in our algorithm can further ensure the inter-onu fairness and intra-onu priority discipline, which is different from other researches. Thus, the lower priority services can be prevented from starvation and the over-allocation problem can be avoided. A performance evaluation shows the effectiveness of the proposed scheme. 2 Network architecture The general multi-wavelength optical access network architecture is shown in Fig. 1. It can realise TPSs including point-to-point service and multicast service. The ring-tree network is comprised of the optical line terminal (OLT), N RNs and N M ONUs. All the remote nodes (RNs) are connected to the OLT in the ring topology and M ONUs are attached to each RN. The OLT is equipped with receiver arrays and two groups of transmitters. Transmitters in Group 1 output downstream signals with N M wavelengths (λ 11, λ 12,,λ nm ) dedicated to each ONU and transmitters in Group 2 provide every RN with shared wavelengths (λ s1, λ s2,,λ sn ) enabling multicast. All the downstream signals are multiplexed through an AWG and then amplified by an erbium-doped optical fibre amplifier (EDFA) and transmitter to RNs. Here, the downstream signals are delivered clockwise and the upstream signals are delivered anticlockwise. This can prevent transmission collision and make access network more reliable. Fig. 2 shows the configuration of RN and ONU. RN consists of circulators, couplers, a wavelength blocker (WB), a 1 (M + 1) AWG and a 1:m splitter. The complex signals for multiple wavelengths are transmitted from one RN to another RN in turn. Especially, the downstream signals are delivered from RN 1 to RN n and the upstream signals are delivered from RN n to RN 1. The transmission route is realised by the configuration of RN. Here, we take RN j, for example, to explain the operation principle. The downstream signals get into RN j through the fibre and they Fig. 1 Multi-wavelength optical access network architecture enabling multicast Fig. 2 Configuration of RN and ONU 1864

3 are routed to the coupler1 by circulator1. Then, the signals are divided into two parts. One part is sent to a WB to enable all wavelengths except the wavelengths (λ sj, λ j1,, λ jm ) to pass, and they will be sent to the feed fibre through circulator 3 for other RNs downstream transmission. The other part is directly sent to an AWG and a set of them (λ 11, λ 12,,λ nm ) will be de-multiplexed for each ONU. We called these wavelengths the dedicated wavelengths. Meanwhile, a shared wavelength (λ sj ) is also de-multiplexed by the AWG and split the multicast signal to every ONU attached to the RN. Both the shared wavelength and the dedicated wavelengths are coupled by CWDM couplers and routed into their corresponding ONUs. In the ONU, the downstream signal is divided into two parts by an optical splitter. One part is received by the two receivers. Receiver1 is for downstream signals on the dedicated wavelength and Receiver2 receives multicast signal on λ sj. The other part is re-modulated and reflected by RSOA for upstream transmission. These upstream signals are sent to the AWG and they are routed to coupler2 by circulator2. Besides, the upstream signals from other RNs are routed to coupler2 by circulator3. Then all the upstream signals are coupled and routed to the ring-based feed fibre by circulator1 for upstream transmission. The direction of upstream transmission is anticlockwise and finally received by the corresponding receivers in the OLT. This architecture features large capacity, high security and easy upgrade by using WDM technology. It enables dynamic wavelength assignment and bandwidths allocation. Here, each ONU is equipped with only two fixed transmitters instead of a tunable transmitter, taking into the account the ONU s installation and equipments cost. What s more, the shared wavelengths in the architecture can not only enable multicast but also realise joint bandwidth allocation by scheduling these shared wavelengths together with the dedicated wavelengths for high utilisation efficiency and fine QoS support. 3 Joint bandwidth allocation on dedicated and shared wavelength for QoS support A joint bandwidth allocation scheme on dedicated and shared wavelength is proposed in this section to efficiently and fairly allocate the bandwidth among all users in the general multi-wavelength optical access network. As mentioned in Section 2, each ONU can employ both the dedicated wavelength and the shared wavelength for bandwidth allocation. In our scheme, the bandwidth allocation is carried out to provide efficient bandwidth utilisation and fine QoS support ensuring inter-onu fairness and intra-onu priority discipline. 3.1 Queue management and priority discipline Since the access network is required to accommodate various kinds of traffic, bandwidth management and fair scheduling of different traffic classes play an important role in supporting QoS. Priority queuing is considered a useful and simple relatively method for supporting differentiated service classes [19, 23, 24]. We classify the traffic into three prioritised services: expedited forwarding (EF) has the highest priority used for strict delay-sensitive services. This is typically a constant-bit-rate voice transmission. Assured forwarding (AF) has medium priority for non-delay-sensitive variable-bit-rate services such as video stream. Best effort (BE) has the lowest priority for delay-tolerant services, which includes web browsing, file transfer and application. The queue management and priority discipline is illustrated in Fig. 3. The EF traffic is stored in EF queue and the AF traffic is stored in AF queue. The BE traffic is most delay tolerant and it is stored in first in, first out (FIFO) queue. The three queues are sent into the weighted round robin (WRR) module to allocate bandwidth according to the weight. The WRR module will satisfy the queue request according to their priority. In other words, AF queue is satisfied right after EF queue and FIFO queue is satisfied last. After intra-onu queue management, OLT schedules the wavelengths and allocate bandwidths for ONUs. Here, a joint bandwidth allocation scheme is proposed. 3.2 Bandwidth allocation on the dedicated wavelength (λ 11, λ 12,, λ nm ) In the general network architecture, each ONU owns a dedicated wavelength for transmission owing to the wavelength-multiplexing technology. In the proposed scheme, bandwidth allocation on the dedicated wavelength is carried out firstly. A REPORT-GATE mechanism is adopted: every ONU reports its bandwidth requests to OLT and OLT grants bandwidth for their upstream transmission according to the corresponding algorithms. Here, we specified the bandwidth allocation in one transmission cycle. The transmission cycle time is the sum of transmission time and guard times for all ONUs. In one transmission cycle time, we assume that each ONU can transmit REPORT message to the OLT. In one transmission cycle, the available bandwidth of wavelength i dedicated to ONUi is expressed as (1). where C i capacity B i available = C i capacity T cycle B control (1) is the of link capacity (bytes/s) of the Fig. 3 Queue management and priority discipline 1865

4 wavelength i, T cycle (s) is the maximum cycle time and B control is the control message length (bytes). The total request bandwidth of ONUi is calculated as follows R i = R i EF + R i AF + R i BE (2) Where R i EF (bytes), Ri AF (bytes) and Ri BE (bytes) are EF, AF, BE request bandwidth of ONUi, respectively. The flowchart of the proposed scheduling scheme is illustrated in Fig. 4. If the total request bandwidth of ONUi R i is not higher than the available ( bandwidth of the dedicated wavelength for ONUi B i ) available, the OLT will allocate bandwidth to each queue according to the request bandwidth; otherwise, the OLT will check the queue length of AF queue firstly. Here, we check AF queue length to ensure fairness among the priority queues. According to the priority discipline mentioned in Section 3.1, we know that higher priority queues can be allocated earlier than lower priority queues. If the higher priority traffic is considerably large, the lower priority traffic will be penalised with indefinite increase in packet delay and loss. So, AF queue length is checked and we set a threshold B th to ensure the efficiency and fairness of the algorithm in the condition of R i is higher than B i available. (Note: the value of B th is determined by iteration to get the optimised one which can obtain the shortest queue delay and prevent buffer overflow.) If the AF queue length is lower than threshold B th, the OLT will compare the request bandwidth of EF queue R i EF with B i available. If the available bandwidth B i available is lower than R i EF, the granted bandwidth for EF queue becomes GEF i = R i EF; otherwise, the OLT will satisfy the request bandwidth of EF queue firstly, and the remaining bandwidth of the wavelength i becomes as B i remain = B i available GEF. i Next, the OLT will compare the request bandwidth of AF queue R i AF with remaining bandwidth B i remain. If the remaining bandwidth B i remain is lower than R i AF, the granted bandwidth for AF queue becomes G i AF = Bi remain ; otherwise, the OLT will satisfy the request bandwidth of R i AF. Finally, the OLT will satisfy the request bandwidth of R i BE and bandwidth granted for BE queue becomes G i BE = B i remain GAF. i This bandwidth allocation process is named Algorithm 1. The pseudo-code of Algorithm 1 is presented as follows in Fig. 5a. If the queue length of AF queue is higher than threshold B th, the granted bandwidth of EF, AF and BE queues are according to the ratio of the available bandwidth B i available, where α, β, γ are the ratio determined in Algorithm 2 in Fig. 5b. It is proved that adaptive fairness among different classes of services is ensured by forcing the ratio policy. Then, the bandwidths granted for three priority queues are given as follows G i EF = a a + b + g Bi available (3) G i b AF = a + b + g Bi available (4) G i g BE = a + b + g Bi available (5) Note that both Algorithm 1 and Algorithm 2 keep to the priority discipline (i.e. bandwidth granted for higher priority queue comes first). This can effectively guarantee the QoS of differentiated services. Moreover, the threshold B th ensures the fairness among priority queues, which prevents EF services from the bandwidth monopolisation. Furthermore, bandwidth can be finely guaranteed according to the priority queues ratio in Algorithm Bandwidth allocation on the shared wavelength (λ sj ) In Algorithm 2, since the bandwidth granted for the three priority queues are based on the ratio α, β and γ, it is likely that there are still remaining EF, AF, BE queues in the ONU s buffer. For EF queue, they are delay-sensitive and bandwidth-guaranteed. So, the buffered EF queues will Fig. 4 Flowchart of the proposed scheduling scheme 1866

5 Fig. 5 Pseudo-code for bandwidth allocation ( a Algorithm 1 R i AF, B ) ( th b Algorithm 2 R i ) AF B th considerably increase the packet delay and delay jitter. Thus, we schedule the shared wavelength to allocate appropriate bandwidth for the transmission of the buffered queues in a transmission cycle. After Algorithm 2, the remained EF, AF, BE queues in the buffer can be calculated as follows Rm i EF = R i a EF a + b + g Bi available (6) bandwidth for EF, AF, BE queues, respectively ( ) G si EF = min Rm i a EF, a + b + g Gi ( ) G si AF = min Rm i b AF, a + b + g Gi ( ) GBE si = min Rm i g BE, a + b + g Gi (10) (11) (12) Rm i AF = R i b AF a + b + g Bi available (7) Rm i BE = R i g BE a + b + g Bi available (8) We suppose the number of the ONUs whose AF queue length is under the threshold (i.e. the number of ONUs who still remains buffered EF, AF and BE queues after bandwidth allocation on the dedicated wavelength in a transmission cycle) is m. Here, the OLT schedule the shared wavelength for the ONUs. The schedule discipline on the shared wavelength is as follows G i = w i m k=1 w i B available (9) In formula (9), G i is the granted bandwidth for ONUi and w i is the weight of each ONU, which is predetermined by service level agreement requirements. We use the predetermined weight of each ONU here for better fairness performance among ONUs. To prevent ONUs from being allocated more bandwidth than requested (i.e. the avoidance of over allocation), we use the following equations to grant It should be pointed out that the bandwidth allocation on the shared wavelength for three priority queues is also according to the queue management and priority discipline mentioned in Section 3.1. Namely, the highest priority queue is served firstly, then the medium priority queue follows and finally the lowest priority queue is served in FIFO way. After aforementioned bandwidth allocation, the bandwidth resources in the network can be fully utilised (including the dedicated wavelength and the shared wavelength). Moreover, the performance of the bandwidth-guaranteed and delay-sensitive services can be improved in terms of average packet delay and packet delay variance. Meanwhile, with the fair-awareness in the scheduling scheme, the lower priority services can be prevented from starvation and the over-allocation problem can be avoided. Thus, the proposed scheme can guarantee the bandwidth and support QoS for differentiated services. Specifically, reduced packet delay and improved jitter performance for EF traffic can be achieved without degrading other traffic. 4 Performance evaluation In this section, the system performance of the proposed bandwidth allocation scheme is evaluated and compared in terms of queue length, throughput rate, end-to-end delay and delay variance. An event-driven packet-based simulation model was developed using C++. We focus on the performance evaluation of the joint bandwidth allocation on dedicated and shared wavelength here rather than the architecture performance. So, we take one RN as 1867

6 an example and the others can be analysed as same as the example. We suppose the total number of ONUs in that RN is 16. Thus, there are 16 dedicated wavelengths and 1 shared wavelength for the 16 ONUs. The link capacity is 1 Gbps. The distance from one ONU to the OLT was assumed to be km and each ONU has a buffer of 10 Mbytes. The guard time is equal to 1 μs and the cycle time is 2 ms. For the traffic model considered here, an extensive study shows that most network services (i.e., http, ftp, variable bit rate (VBR) video applications, etc.) can be characterised by self-similarity and long-range dependence (LRD) [23, 24]. This model was adopted to generate highly bursty BE and AF traffic classes with the Hurst parameter of 0.7. The packet sizes were uniformly distributed between 64 and 1518 bytes. Additionally, high priority traffic (e.g. voice applications) was modelled by a Poisson distribution and the packet size was fixed to 70 bytes. Fig. 6 plots the average queue length in each of the three priority queues in the buffer of the ONU. Note that the EF, AF and BE traffic loads were specified as 20, 30 and 50% of the network offered load, respectively. Comparing the three traffic classes, it can be seen that the BE traffic has the longest queue length followed by the AF traffic and the EF traffic. We can find that the BE traffic reaches its stability limit at a network offered load of about 1.6 Gbps, whereas the AF and EF traffic remain in a stable condition. By observing the average queue length of the three priority queues, we can obtain the value of B th in the proposed bandwidth allocation scheme. The value of B th is optimised by iteration to obtain the shortest queue delay and prevent buffer overflow. In this simulation, we get the value of B th as 56% of the total queue length in the buffer. When the threshold B th is determined, the joint bandwidth allocation on dedicated and shared wavelength can be carried out according to Fig. 4 in Section 3. Fig. 7 shows the comparison of the average throughput rate under different numbers of ONUs on the shared wavelength. In our scheme as mentioned in Section 3, if the bandwidth of the dedicated wavelength cannot satisfy the bandwidth requirement of the corresponding ONU, then the shared wavelength can be scheduled to allocate bandwidth for that ONU. In other words, not every ONU needs bandwidth allocation on the shared wavelength. Only those who have heavy traffic will employ the shared wavelength. Here, we denote the number of ONUs who employ the shared wavelength as N s (i.e. the number of ONUs with heavy Fig. 6 Average queue length of three priority queues Fig. 7 Comparison of average throughput rate traffic isn s ). In Fig. 7, N s = 0 represents the condition of the WDM access optical network with no shared wavelength. Fig. 7 shows the throughput rates under different N s are almost the same when the offered load is less than 0.4. This is because when the offered load is lower, the dedicated wavelength can fully satisfy the bandwidth requirement of the ONUs and there is no need to allocate bandwidth on the shared wavelength. However, for higher load, Fig. 7 shows that the proposed scheme offered as high as 74.8% throughput, compared with 56.6% when there is no shared wavelength employed. The reason for that is the joint bandwidth allocation scheme greatly eases the burden of the dedicated wavelength and fully utilises bandwidth of the shared wavelength. And the more ONUs are allocated bandwidth on the shared wavelength, the higher throughput they can obtain. This is achieved by eliminating more of the BE traffic starvation station in ONU buffers. Fig. 8 compares the average end-to-end packet delay and the end-to-end delay of the EF traffic classes under different numbers of ONUs on the shared wavelength. As Fig. 8a illustrated, the proposed scheme with joint bandwidth allocation on the shared wavelength (i.e. N s =4, 8, 16) reduces the average end-to-end delay compared with the situation of N s = 0 (i.e. with no shared wavelength). This shows the advantage of our proposed scheme that is the capacity of assigning ratio (α, β, γ) in a way to poise the performance of each traffic aside. Also, it is obvious that with an increase in the traffic amount, the average end-to-end delay increases because both the mechanisms (with or without the shared wavelength) will improve high-priority traffic end-to-end delay but sacrifice the low-priority traffic delay performances to some extent according to our priority queue discipline. Furthermore, the simulation results show that the shortest average end-to-end delay is achieved when N s = 12. It is because the queue delay of the traffic especially for BE traffic will be considerably reduced when the more ONUs employ the shared wavelength. Fig. 8b shows that the joint allocation scheme with shared wavelength can reduce the EF end-to-end delay. With the shared wavelength in our scheme, the EF traffic end-to-end delay remains almost the same (only ms) unless the offered load exceeds 50%. The reason is that bandwidth for 1868

7 Fig. 8 End-to-end delay a Average end-to-end delay b EF end-to-end delay the EF traffic is always allocated firstly and the bandwidth requirement of EF traffic can be satisfied as soon as possible under the lower offered load. But when the load is higher, the queue delay in the heavy queue state inevitably increases. Thus, the EF end-to-end delay increases. Furthermore, we can find out that with more ONUs employing the shared wavelength, the delay decreases unless the load exceeds 90%. It is because the buffered time in the ONUs can be greatly reduced by scheduling the shared wavelength. But when the load is considerably high (as 90% or higher), the increasing number of ONUs on the shared wavelength will contend for bandwidth and bring about delay. At the same time, by enforcing the ratio policy and priority discipline, our scheme make sure to provide each PQ a share of the assigned bandwidth. In this way, fairness is ensured among different classes of traffic, and the traffic priority (delay sensitivity) is respected. Fig. 9 shows the comparison of the jitter performance for EF traffic under different numbers of ONUs on the shared wavelength. The delay variance σ 2 is calculated as s 2 = N 1 ( d ) 2/N d EF i where d EF i represents the delay time of the EF packet i, d represents the average packet delay and N is the total number of received EF packets. Simulation result shows that the delay variance is decreased when the offered load rises. The reason is the EF traffic is always transmitted ahead of other priority traffic to reduce transmission jitter. Moreover, when the offered load increases, the amount of the EF traffic also increases. Then, the network status tends to stabilise owing to the constant-bit-rate property of the EF traffic. Meanwhile, when the traffic load increases the cycle time tends to stabilise, which will lead to the improvement of the jitter performance. Moreover as expected, our scheme of joint bandwidth allocation on the dedicated and shared wavelength achieves improved jitter performance compared with the bandwidth allocation scheme with no shared wavelength as illustrated in Fig. 9. The reason is our scheme efficiently allocates the remaining bandwidth of the shared wavelength to all priority queues and it prevents bandwidth monopolisation by checking the AF queue length. 5 Conclusion In this paper, a joint bandwidth allocation scheme on dedicated and shared wavelength is proposed to effectively utilise the network bandwidth resources in multi-wavelength optical access network. The proposed scheme schedules the shared wavelength that is used for multicast as well as the dedicated wavelength for the ONUs, thus enhancing system performances in terms of the end-to-end packet delay and system throughput. Moreover, the priority queue discipline is adopted here to support QoS. The simulation results confirm that our joint bandwidth allocation scheme can reduce packet delay and packet delay variation for high-priority services to ensure QoS. Furthermore, the threshold and ratio in our algorithm can further ensure the inter-onu fairness and intra-onu priority discipline. Thus, the lower priority services can be prevented from starvation and the over-allocation problem can be avoided. Fig. 9 EF delay variance 6 Acknowledgments This work was supported by the Programmes of Natural Science Foundation of China (grant numbers and ), Shanghai Science and Technology Development Funds (grant numbers and ), Shanghai Leading Academic Discipline Project and STCSM (grant numbers S30108 and 08DZ ). 1869

8 7 References 1 Iwatsuki, K., Kani, J.-i., Suzuki, H., Fujiwara, M.: Access and metro networks based on WDM technologies, J. Lightwave Technol., 2004, 22, (11), pp Chang, G.-K., Chowdhury, A., Jia, Z., et al.: Key technologies of WDM-PON for future converged optical broadband access networks, J. Opt. Commun. Netw., 2009, 1, (4), pp. C35 C50 3 Bock, C., Arellano, C., Prat, J.: Resilient single-fiber ring access network using coupler-based OADMs and RSOA-based ONUs. Optical Fiber Communication Conf. and National Fiber Optic Engineers Conf., California, America, March Esmail, M.A., Fathallah, H.: Fiber fault management and protection solution for ring-and-spur WDM/TDM long-reach PON. IEEE Global Telecommunications Conf., Texas, USA, 2011, pp An, F.-T., Soo Kim, K., Gutierrez, D., et al.: SUCCESS: a next generation hybrid WDM/TDM optical access network architecture, J. Lightwave Technol., 2004, 22, (11), pp Zhang, Y., Deng, N., Chan, C.-K., et al.: A multicast WDM-PON architecture using DPSK/NRZ orthogonal modulation, IEEE Photonics Technol. Lett., 2008, 20, (17), pp Lingzhi, G., Shilin, X., Zhixin, L., et al.: A scheme to realize multicast/ broadcast by superimposing DPSK signal onto Manchester/NRZ signal. Communications and Photonics Conf. and Exhibition, Shanghai, China, 2009, pp Han, K.-E., Yang, W.-H., Yoo, K.-M., Kim, Y.-C.: Design of AWG-based WDM-PON architecture with multicast Capability. IEEE INFOCOM Workshops 2008, Phoenix, AZ, pp Kim, N.U., Kang, M.: Traffic share-based multicast scheduling for broadcast video delivery in shared-wdm-pons, J. Lightwave Technol., 2007, 25, (9), pp Kramer, G., Mukherjee, B., Pesavento, G.: Interleaved polling with adaptive cycle time (IPACT): A dynamic bandwidth distribution scheme in an optical access network, Photonics Netw. Commun., 2002, 4, (1), pp McGarry, M., Reisslein, M., Maier, M.: Ethernet passive optical network architectures and dynamic bandwidth allocation algorithms, IEEE Commun. Surv. Tutor., 2008, 10, (3), pp Zheng, J., Mouftah, H.: A survey of dynamic bandwidth allocation algorithms for ethernet passive optical networks, Opt. Switch. Netw., 2009, 6, (3), pp Ma, M., Zhu, Y., Cheng, T.: A bandwidth guaranteed polling MAC protocol for Ethernet passive optical networks. Proc. IEEE INFOCOM, San Francisco, CA, March 2003, vol. 1, pp Hwang, J., Yoo, M.: QoS-aware class gated DBA algorithm for the EPON system. Int. Conf. Advanced Technologies for Communications, Hanoi, October 2008, pp Chen, J., Chen, B., Wosinska, L.: Joint bandwidth scheduling to support differentiated services and multiple service providers in 1 G and 10 G EPONs, J. Opt. Commun. Netw., 2009, 1, (4), pp Berisa, T., Ilic, Z., Bazant, A.: Absolute delay variation guarantees in passive optical networks, J. Lightwave Technol., 2011, 29, (9), pp Okumura, Y.: Traffic control algorithm offering multi-class fairness in PON based access networks, IEICE Trans. Commun., 2010, 93, (3), pp Hwang, I.-S., Lee, J.-Y., Robert Lai, K., Liem, A.T.: Generic QoS-aware interleaved dynamic bandwidth allocation in scalable EPONs, J. Opt. Commun. Netw., 2012, 4, (2), pp Sherif, S.R., Hadjiantonis, A., Ellinas, G., et al.: A novel decentralized ethernet-based PON access architecture for provisioning differentiated QoS, J. Lightwave Technol., 2004, 22, (11), pp Kiaei, M.S., Assi, C., Meng, L., Maier, M.: On the co-existence of 10 G-EPON and WDM PONs: a scheduling and bandwidth allocation approach, J. Lightwave Technol., 2011, 29, (10), pp Hara, K., Nakamura, H., Kimura, S., et al.: Flexible load balancing technique using dynamic wavelength bandwidth allocation (DWBA) toward 100 Gbit/s-class-WDM/TDM PON. ECOC 2010, Torino, Italy, September, 2010, pp Zhu, M., Guo, W., Xiao, S., et al.: Design and performance evaluation of dynamic wavelength scheduled hybrid WDM/TDM PON for distributed computing applications, Opt. Express, 2009, 17, (2), pp Blake, S., Black, D., Carlson, M., Davies, E., Wang, Z., Weiss, W.: An architecture for differentiated services. IETF RFC 2475, Assi, C., Ye, Y., Dixit, S., Ali, M.A.: Dynamic bandwidth allocation for quality of service over Ethernet PONs, IEEE J. Sel. Areas Commun., 2003, 21, (9), pp