Abstract

In this paper we present and analyze the model for wavelength and bandwidth allocation in hybrid TDM/WDM EPONs with full quality of service (QoS) support, called the dynamic wavelength priority bandwidth allocation with fine scheduling (DWPBA-FS). In order to implement the QoS support, we present a new approach for QoS analysis and implementation in WDM EPON, an approach in which wavelength assignment takes place per service class and not per ONU, which is a case that has widely been suggested by the common approach in literature.

© 2011 OSA

1. Introduction

Until today, different models for incremental migration from TDM to TDM/WDM EPON networks have been proposed, but most attention has been given to those solutions that support quality of service (QoS) implementation. Moreover, with the rapid development of different bandwidth applications, QoS support is becoming a key concern in WDM EPON network just as it was similarly the case with EPON networks. Under this assumption, a new concept for optical access networks will be needed. The presented dynamic wavelength priority bandwidth allocation with fine scheduling (DWPBA-FS) model implements a novel approach in analyzing QoS in WDM EPON in which wavelength assignment takes place per service class and not per ONU, as it has been a common approach in literature so far, is used [1]. In this way, we avoid the need for the implementation of additional complex algorithms for QoS support which in turn significantly reduces dynamic bandwidth algorithm (DBA) complexity and directly decreases system cost in comparison with various QoS DBA models which have been published until now. The presented model includes a hybrid inter/intra-ONU scheduling mechanism where both the optical line terminal (OLT) and the optical network unit (ONU) are responsible for performing packet scheduling. Furthermore, the model includes an extension of the multipoint control protocol (MPCP) allowing the presented architecture and model an incremental upgrade from TDM EPON to TDM/WDM EPON based on need. Moreover, with multiple wavelength assignment for each ONU, it offers great flexibility and allows service providers to respond to the increase of user demand in terms of bandwidth and other QoS parameters in a very simple and cost-effective manner. This ‘on-demand’ concept and system capability to handle the diversity in user demand is currently considered as a key factor for the realization of the new multiservice access network [2].

Detailed simulation experiments are presented with the aim of conducting a study of the performances and obtaining the validation of effectiveness of the presented solution. The presented results show that DWPBA-FS model outperforms dynamic wavelength and bandwidth allocation (DWBA) models which have been suggested so far. Moreover, a comparison with single-channel EPON with QoS support is given in order to emphasize the increase in efficiency and underline the necessity of WDM technology implementation in EPONs for the successful realization of the new generation optical access networks.

2. DWBA-FS model

The proposed WDM EPON system is actually a tree-based EPON with support to multiple wavelengths (λ1, λ2, λ3, λ4) in every ONU and OLT. In the initial phase of migration, we analyze a solution with an array of fixed-tuned transceivers in the OLT and ONUs, one for each operating wavelength channel, in order to enable simultaneous transmission of traffic in one station on different wavelengths. Furthermore, only units with higher traffic demands may be WDM upgraded. Namely, with an array of fixed-tuned transceivers, every station is able to simultaneously use four different wavelengths and exchange REPORT and GATE MPCP control messages.

The presented model is based on the DWPBA (Dynamic Wavelength Priority Bandwidth Allocation) model [1], where the DWPBA-FS model further develops the presented concept with the introduction of new traffic subclasses in the system in accordance with Diffserv model [3]. Furthermore, in the DWPBA-FS model, the control wavelength used exclusively for synchronization and exchange control messages in the basic model, is now used for data transmission. Exchange of control messages in the enhanced model now takes place on data wavelengths, where the processing is carried out in accordance with the MPCP protocol. In order to support differentiated services over WDM EPON, each ONU is installed with an array of physical queues where each queue is used to store a defined class of traffic. When a packet arrives to ONU, it will be first categorized into one of the priority groups according to its content and then be placed into one of the queues. Accordingly, the network traffic should be categorized into different classes: EF (Expedited Forwarding) - highest priority traffic class for delay-sensitive traffic with constant bit rate which is intended for services such as voice and other delay-sensitive applications that require bounded end-to-end delay and jitter specifications; AF (Assured Forwarding) - medium priority traffic for not delay-sensitive traffic with variable bit rate. AF class is intended for services, such as video transmission, that are not delay-sensitive but which require bandwidth guarantees. Assured forwarding allows the operator to provide assurance of delivery as long as the traffic does not exceed a certain subscribed rate. Traffic that exceeds the subscription rate faces a higher probability of being dropped if congestion occurs. The AF traffic class is further divided into four subclasses according to the standard (traffic is listed by priority, from highest to the lowest): AF4 (e-commerce applications), AF3 (mission critical application), AF2 (non-organization streaming audio and video), and AF1 (bulk traffic) [3]. The first two AF subclasses could be defined as premium and normal business applications, and another two as premium and home business applications. If the congestion occurs between classes, the traffic in the higher subclass is given priority over lower traffic subclasses; BE (Best-Effort) - low priority traffic class for delay tolerable services that do not require any guarantees in terms of jitter and bandwidth.

In order to reserve bandwidth for the four AF subclasses, we propose the implementation of the Weighted Fair Queuing (WFQ) as a scheduling algorithm in ONUs since this method is able to automatically smooth out the flow of data by sorting packets that minimize the average latency and improve system performances. Also, WFQ can be described as a queuing algorithm that combines fair queuing and preferential weighting. The fairness aspect of WFQ behaves similarly to round-robin queuing, with queues serviced in a continuously repeating sequence from top to bottom, and then starting at the top again. The weighting aspect of WFQ applies a “weight” to a queue that indicates the importance of the queue in relation to the available resources. The weight is used to ensure that more important queues are processed more often than other less important queues. With WFQ, queues are first sorted in order against of their increasing weighted value. Then, each queue is serviced according to its weighted proportion to the available resources. By using WFQ, each priority group can reserve different weighted proportions in the next transmission window and the packet scheduling mechanisms allow different sessions to have different service shares [4]. Therefore, when the next GATE message arrives, the ONU can transmit packets from priority queues up to the amount that has previously been reserved. Accordingly, we adopted the following scheme for transmission and segregation of different traffic classes by wavelengths: λ1 is allocated for transmission of EF traffic class, λ2 and λ3 are allocated for transmission of AF traffic class, i.e. AF1, AF2, AF3 and AF4 traffic subclasses, λ4 is allocated for transmission of BE traffic class. For the transmission of medium priority subclasses, we suggest the use of two wavelengths in one ONU simultaneously since we assume that AF traffic class is the most present in the system. Namely, according to the new traffic scheme in modern networks, most applications and services are multimedia-based and in years to come, the different forms of video applications will account for approximately 90% of consumer traffic [5]. On the other hand, EF traffic class, although being the highest priority class, includes voice applications that are not bandwidth intensive (10% to 20% of total generated traffic in the system) and with the allocation of one wavelength solely for its transmission, even the most rigorous QoS requirements can be met. BE traffic class generally occupies more bandwidth in comparison to EF class but, as we have previously explained, applications in this class do not require any QoS guarantees. Our proposed model ensures fairness among all traffic classes, especially for BE traffic which, in most of the ”to-date” public schemes, pays the performance penalty to gratify higher priority traffic. With the allocation of one wavelength for this class, transmission characteristic of this class will be significantly improved in relation to different models which have been suggested so far. Having segregated the transmission of traffic classes in the described manner, there is no need for the implementation of an additional algorithm for QoS support.

The presented model is completely in compliance with the IEEE 802.3ah standard and introduces offline scheduling in order to support the QoS implementation. Scheduling in ONUs is defined in the following manner. Each ONU will follow a request-per-class trend to report its bandwidth requirements of each traffic class separately. Hence, each ONU will send three REPORT messages: one for EF traffic (λ1), one for all AF subclasses (λ2), and one for BE class (λ4). Figure 1(a) illustrates the proposed intra-ONU scheduler. Since ONUs have the ability to transmit packets on λ2 and λ3 simultaneously, WFQ-based scheduler will first schedule packets from Q4 and Q3 queues in accordance with allocated bandwidth and defined weight factors. At the same time, packets from lower priority queues (Q2 and Q1) will be scheduled for transmission on λ3. In case the transmission window (TW) is still available and Q4 and Q3 queues are empty or in case the remaining packets could not fit in TW (fragmentation is not allowed), scheduler will try to schedule packets that belong to the AF2 and/or AF1 traffic classes in the remaining TW and vice versa. In this way the scheduling mechanism efficiently uses system resources and adjusts the current distribution of traffic in the system in accordance with user demands.

Communication and data exchange between OLT and ONUs is defined as follows. In the downstream direction, by means of the broadcasting mechanism, the OLT sends data to the ONUs simultaneously on multiple wavelengths, which are in turn received by the ONUs based on the destination MAC address (as is the case with the conventional EPON). However, in the upstream direction, apart from the bandwidth allocation, the allocation of resources now also includes the wavelength allocation. At the beginning of the cycle, OLT allocates four data wavelengths to the selected ONUs based on the received bandwidth requests and the mathematical model described below. When one ONU finishes transmission on allocated wavelength(s), that wavelength(s) is (are) automatically shifted to another ONU for transmission of traffic that belongs to the same class according to decision made in OLT, Fig. 1(b). Accordingly, since in one cycle every ONU must receive an opportunity to transmit traffic of each traffic class, OLT must maintain information about allocated wavelengths and ONUs which used them. As wavelength allocation is fixed, OLT has to make a decision only about bandwidth allocation, which significantly simplifies DBA complexity and improves system performances. The DBA algorithm used in the DWPBA-FS model is based on the modified gated IPACT (MG-IPACT) algorithm presented in [1]. This scheme will grant ONU a transmission window whatever size it previously requested and in accordance with the weight factors associated with every ONU for each supported traffic class. Thus, the largest possible granted window size will be the maximum length of ONU’s queue. In the presented mathematical model, we consider WDM EPON with N ONUs. The transmission speed of the EPON is R Mbps. The total available upstream bandwidth of each wavelength in one cycle is:

Wtotal=R(Tmax_cycleN*Tg)

Allocated bandwidth for all three traffic classes for ONUi can be calculated as:

Wtotaltc_requested=iNWitc_requested,tc{EF,AF,BE},
witc=Witc_requestedWtotaltc_requested,tc{EF,AF,BE}andi=1Nwitc=1.,
Witc_allocated={Wqueue,Witc_requestedWqueuewitc*Wtotal,Witc_requested<Wqueue,tc{EF,AF,BE},
where we denote Tmax_cycle as the maximum granted transmission cycle time during which all ONUs can transmit data or/and send REPORT messages to the OLT and is limited to 2 ms [6]. We also denote Tg as the guard time that separates the transmission windows of two consecutive ONUs and witc as the weight assigned to each ONU for each supported traffic class. Further, Witc_requesteddenotes the amount of bandwidth that ONUi requests for the transmission of supported traffic classes, Wtotaltc_requesteddenotes the total requested bandwidth of all ONUs for each traffic class, Witc_allocateddenotes the amount of bandwidth that OLT allocates for the transmission of each traffic class in ONUi, Wqueue denotes the maximum defined length of ONUs queues.

3. Simulation results and performance evaluation

The presented model is tested using a WDM EPON network model developed in Matlab, using Simulink packet. In our simulations, we consider a WDM EPON consisting of 64 ONUs that support traffic transmission on four different wavelengths in each station and speed of each wavelength is set to 1 Gbps. In the simulation, we implement standardized parameters widely used in the literature for the simulation of EPONs [1, 7]. As we have previously explained, AF traffic class is now dominant [5, 6] hence we adopt the following traffic profile: 15% of the total generated traffic is considered as EF, 50% as AF and the remaining 35% is considered to be BE traffic class. Weighted proportions used in WFQ mechanism are 0.2, 0.1, 0.5, and 0.2 for Q4, Q3, Q2 and Q1, respectively. The communication quantities used for comparison are: average packet delay, jitter, throughput rate, and packet loss rate.

First we analyze average packet delay given in Fig. 2(a) . Simulation results follow the load distribution between AF subclasses and consequently AF3 subclass which is the least represented in the system and has the best characteristic of the delay, while the most presented AF2 subclass has the largest delay. However, with the implementation of two wavelengths for AF class transmission, average packet delay for this class decreases below 1.7 ms (a decrease of 35.5% compared with the DWPBA model tested under the same conditions and with 64 ONUs) which would allow service provider to guarantee efficient delivery of multimedia traffic. At the same time, the characteristics of EF and BE class are not significantly degraded given the fact that the DWPBA-FS model abolishes a separate wavelength for synchronization and shift control functions to data wavelength. Average packet delay for EF and BE classes has increased (negligible degradation of 1.7% and 2.1% in comparison with the DWPBA model) [1]. However, the presented results confirm the excellent performance of the DWPBA-FS model because the average packet delays of all traffic classes amount to less than 2.3 ms, which enables the QoS provision even for the low priority traffic class. Further, we analyze EF packet delay variation as the essential QoS parameter for the assessment of network performance. The jitter is represented by the packet delay variation of two consecutively departed EF packets from the same ONU in the same transmission window [1, 7]. Figure 2(b) shows the probability density function (pdf) of EF service packet delay at full loading scenario for single-channel hybrid grant protocol with priority based scheduling (HG (PBS)) [7], DWPBA and DWPBA-FS models. EF delay sequence presents dispersion with enough number of data points in a tail until 2.1 ms for HG (PBS), 1 ms for DWPBA, and 1.2 ms for DWPBA-FS (negligible degradation of only 0.2 ms in comparison with DWPBA). Presented results confirm that the DWPBA-FS model is able to provide an excellent EF jitter and show superiority of multichannel EPON over single-channel systems. Namely, for the successful transmission of EF traffic class through the network, it is vital that the value of jitter be as low as possible. This is essential for the definition of QoS policies and service level agreements (SLAs) in multimedia-based networks in which voice transmission exists since voice-based applications are particularly sensitive to delay and delay variation. The high performances of the DWPBA-FS model are also confirmed with the comparison of the packet loss rate and network throughput, Fig. 2(c). Packet loss in the system mainly comes from the fact that packet fragmentation is not allowed and therefore large packets that do not fit in the currently granted window will have to be postponed and consequently lost. AF traffic class is the most present traffic class and is often characterized with large multimedia packets (streaming, multimedia transmission) and therefore need fragmentation. Hence, the efficient transmission of this traffic class is essential for low packet loss. With the implementation of AF subclasses, WFQ mechanism, and the two wavelengths for transmission, packet loss is significantly decreased. The DWPBA-FS model improves system performances since it increases the probability that a large packet could be transmitted immediately. At the highest network load, packet loss in the DWPBA-FS is decreased by 3.2 times in comparison with the HG (PBS) and by 30% in comparison with the DWPBA model. Consequently, minimized packet loss led to maximized network throughput, Fig. 2(c) (Inserted). Single-channel HG (PBS) uses only one wavelength for upstream transmission (the maximum load in the singlechannel EPON is four times smaller in comparison with the 4-channels WDM EPON (1 Gbps vs. 4 Gbps)) hence characteristic rapidly enters the saturation while DWPBA has better characteristics than HG (PBS) but lower efficiency in comparison with the DWPBA-FS. Figure 2(d) shows comparison of the throughput characteristics of the dominant AF traffic class (one wavelength in the DWPBA, two wavelengths in the DWPBA-FS model). The DWPBA (AF) characteristic first comes into saturation (at load of 0.45), while the DWPBA-FS model distributes load on two wavelengths and at the highest load, AF throughput is doubled. This analysis confirms that the DWPBA-FS model increases the bandwidth utilization and overall system efficiency. Moreover, it justifies the introduction of two wavelengths for the transmission of AF traffic and confirms the model superiority.

 

Fig. 2 (a) Average packet delay in DWPBA-FS model. (b) Comparison of jitter performances.(c) Comparison of packet loss rate; Inserted: Comparison of network throughput. (d) AF throughput rate.

Download Full Size | PPT Slide | PDF

4. Conclusion

The DWPBA-FS model presented in this work introduces novel intra/inters ONU scheduling mechanisms for the WDM EPONs. The model features low implementation complexity, hardware implementation ability, and could be implemented on a per need basis. Extensive simulation results confirm that the proposed scheme maintains the average packet delay and jitter performance of high priority traffic and at the same time achieves a very fine degree of AF traffic segregation. The presented model improves packet loss rate and network throughput and is able to guarantee efficient transmission of AF traffic subclasses, i.e. definition of SLAs for the transmission of business and multimedia applications, which is a key factor required for the successful on-field model implementation. Moreover, the proposed novel concept reduces overall system cost and offers the superior network scalability and flexibility needed for meeting various user demands which makes it suitable for implementation in the multiservice new generation optical access networks.

Acknowledgments

This work was supported by the Serbian Ministry of Education and Science under contract No. 171011.

References and links

1. M. Radivojević and P. Matavulj, “Novel wavelength and bandwidth allocation algorithms for WDM EPON with QoS support,” Photonic Netw. Commun. 20(2), 173–182 (2010). [CrossRef]  

2. H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

3. S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, and W. Weiss, An Architecture for Differentiated Services, RFC 2475 (IETF, 1998).

4. K. Kwong, D. Harle, and I. Andonovic, “Dynamic bandwidth allocation algorithm for differentiated services over WDM EPONs,” in Proceedings of 9th International Conference on Communications Systems (IEEE, 2004), pp. 116–120.

5. Cisco systems white paper, “Forecast and Methodology, 2010–2015” (Cisco Systems, 2011). http://www.cisco.com/en/US/netsol/ns827/networking_solutions_white_papers_list.html

6. F. J. Hens and J. M. Caballero, Triple Play: Building the converged network for IP, VoIP and IPTV (John Wiley & Sons, 2008), Chaps. 1, 6–9.

7. M. Radivojevic and P. Matavulj, “Implementation of Intra-ONU scheduling for quality of service support in ethernet passive optical networks,” J. Lightwave Technol. 27(18), 4055–4062 (2009). [CrossRef]  

References

  • View by:
  • |
  • |
  • |

  1. M. Radivojević and P. Matavulj, “Novel wavelength and bandwidth allocation algorithms for WDM EPON with QoS support,” Photonic Netw. Commun. 20(2), 173–182 (2010).
    [CrossRef]
  2. H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).
  3. S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, and W. Weiss, An Architecture for Differentiated Services, RFC 2475 (IETF, 1998).
  4. K. Kwong, D. Harle, and I. Andonovic, “Dynamic bandwidth allocation algorithm for differentiated services over WDM EPONs,” in Proceedings of 9th International Conference on Communications Systems (IEEE, 2004), pp. 116–120.
  5. Cisco systems white paper, “Forecast and Methodology, 2010–2015” (Cisco Systems, 2011). http://www.cisco.com/en/US/netsol/ns827/networking_solutions_white_papers_list.html
  6. F. J. Hens and J. M. Caballero, Triple Play: Building the converged network for IP, VoIP and IPTV (John Wiley & Sons, 2008), Chaps. 1, 6–9.
  7. M. Radivojevic and P. Matavulj, “Implementation of Intra-ONU scheduling for quality of service support in ethernet passive optical networks,” J. Lightwave Technol. 27(18), 4055–4062 (2009).
    [CrossRef]

2010 (2)

M. Radivojević and P. Matavulj, “Novel wavelength and bandwidth allocation algorithms for WDM EPON with QoS support,” Photonic Netw. Commun. 20(2), 173–182 (2010).
[CrossRef]

H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

2009 (1)

Iiyama, N.

H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

Kimura, H.

H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

Kumozaki, K.

H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

Matavulj, P.

M. Radivojević and P. Matavulj, “Novel wavelength and bandwidth allocation algorithms for WDM EPON with QoS support,” Photonic Netw. Commun. 20(2), 173–182 (2010).
[CrossRef]

M. Radivojevic and P. Matavulj, “Implementation of Intra-ONU scheduling for quality of service support in ethernet passive optical networks,” J. Lightwave Technol. 27(18), 4055–4062 (2009).
[CrossRef]

Radivojevic, M.

M. Radivojević and P. Matavulj, “Novel wavelength and bandwidth allocation algorithms for WDM EPON with QoS support,” Photonic Netw. Commun. 20(2), 173–182 (2010).
[CrossRef]

M. Radivojevic and P. Matavulj, “Implementation of Intra-ONU scheduling for quality of service support in ethernet passive optical networks,” J. Lightwave Technol. 27(18), 4055–4062 (2009).
[CrossRef]

Sakai, Y.

H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

IEICE Trans. Commun. (1)

H. Kimura, N. Iiyama, Y. Sakai, and K. Kumozaki, “A WDM based future optical access network and support technologies for adapting the user demands’ diversity,” IEICE Trans. Commun. E93-B, 246–254 (2010).

J. Lightwave Technol. (1)

Photonic Netw. Commun. (1)

M. Radivojević and P. Matavulj, “Novel wavelength and bandwidth allocation algorithms for WDM EPON with QoS support,” Photonic Netw. Commun. 20(2), 173–182 (2010).
[CrossRef]

Other (4)

S. Blake, D. Black, M. Carlson, E. Davies, Z. Wang, and W. Weiss, An Architecture for Differentiated Services, RFC 2475 (IETF, 1998).

K. Kwong, D. Harle, and I. Andonovic, “Dynamic bandwidth allocation algorithm for differentiated services over WDM EPONs,” in Proceedings of 9th International Conference on Communications Systems (IEEE, 2004), pp. 116–120.

Cisco systems white paper, “Forecast and Methodology, 2010–2015” (Cisco Systems, 2011). http://www.cisco.com/en/US/netsol/ns827/networking_solutions_white_papers_list.html

F. J. Hens and J. M. Caballero, Triple Play: Building the converged network for IP, VoIP and IPTV (John Wiley & Sons, 2008), Chaps. 1, 6–9.

Cited By

OSA participates in CrossRef's Cited-By Linking service. Citing articles from OSA journals and other participating publishers are listed here.

Alert me when this article is cited.


Figures (2)

Fig. 1
Fig. 1

ONU scheduling.

Fig. 2
Fig. 2

(a) Average packet delay in DWPBA-FS model. (b) Comparison of jitter performances.(c) Comparison of packet loss rate; Inserted: Comparison of network throughput. (d) AF throughput rate.

Equations (4)

Equations on this page are rendered with MathJax. Learn more.

W total =R( T max_cycle N* T g )
W total tc_requested = iN W i tc_requested ,tc{ EF,AF,BE } ,
w i tc = W i tc_requested W total tc_requested ,tc{ EF,AF,BE }and i=1 N w i tc =1. ,
W i tc_allocated ={ W queue , W i tc_requested W queue w i tc * W total , W i tc_requested < W queue , tc{ EF,AF,BE } ,

Metrics