Open Access
Add to BibSonomy
Abstract
We propose an adaptive wavelength allocation pattern for scheduling multi-wavelength ONUs in the NG-EPON. Proposed wavelength allocation patterns make online decisions for ONU wavelength allocation based on an adaptive threshold that reflects both ONU’s absolute bandwidth request size as well as the relative bandwidth request size to other ONUs. The proposed wavelength allocation pattern has low complexity and good network performance. Simulation results show that the proposed wavelength allocation pattern achieves small packet delay, huge bandwidth throughput, and a low packet loss ratio in different network conditions, which outperforms the existing wavelength allocation patterns for the NG-EPON.
© 2018 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
To meet the continuously increasing bandwidth demand of access networks, IEEE initializes a working group to study the Next Generation Ethernet Passive Optical Network (NG-EPON), which can provide huge access data rate [1]. Currently, the NG-EPON 802.3ca task group focuses on the standardization of 25G and 50G EPON. In the NG-EPON, each wavelength channel works at the rate of 25 Gb/s. In order to achieve higher ONU peak rate, channel bonding that coordinates multiple physical wavelength channels is introduced [2]. With channel bonding, NG-EPON ONUs are capable of working on multiple wavelength channels simultaneously. As Fig. 1 shows, NG-EPON supports single-wavelength 25G ONUs (e.g., ONU #1) as well as channel-bonded multi-wavelength 50G ONUs (e.g., ONU #m). Specifically, 25G ONUs work on the wavelength λ1 while 50G ONUs work on wavelength λ1 and λ2. Similar to conventional 1G-EPON [3] and 10G-EPON [4] ONUs, NG-EPON ONUs share the total available bandwidth in time-division multiplexing (TDM) way in the upstream. The upstream bandwidth negotiation between the Optical Line Terminal (OLT) and ONUs is performed in a request-and-grant way. ONUs send REPORT messages to report their backlogged data sizes to the OLT. The Dynamic Wavelength and Bandwidth Allocation (DWBA) in the OLT will calculate and assign transmission wavelengths and timeslots for each ONU. Then the OLT will inform ONUs of DWBA results by sending GATE messages to them. ONUs can only transmit their data during assigned timeslots on allocated wavelengths as indicated in GATE messages.
NG-EPON multi-wavelength ONUs can be assigned to transmit their data on either single wavelengths or multiple wavelengths flexibly. As Fig. 1 shows, a 50G ONU can transmit its data on single wavelength with a period of time (e.g. ONU #m) while it can also transmit the same amount of data on two wavelengths simultaneously with one half of the former time (e.g. ONU #n). The total amount of allocated bandwidth remains the same and the only difference is the number of transmission wavelengths. Intuitively, transmission on multiple wavelengths has earlier finish time while the data distribution and aggregation among multiple wavelengths needs extra processing due to necessary demultiplexing and multiplexing mechanism [5]. Transmission on single wavelength is easier to implement while it needs longer transmission time. Moreover, transmission on multiple wavelengths needs more guard time while transmission on single wavelength only needs one, which will increase the average delay. Since NG-EPON ONUs can have various wavelength allocation patterns, i.e., transmitting on single or multiple wavelengths, the key issue is whether there exist huge differences between these wavelength allocation patterns. It is very important to figure out the influence of wavelength allocation patterns (single or multiple wavelengths) on network performances, which will be conducive to design and implementation of DWBA algorithms in NG-EPON. Furthermore, efficient and comprehensive wavelength allocation patterns that have excellent network performances such as small packet delay and large bandwidth throughput need to be studied.
In this paper, we investigate different wavelength allocation patterns and propose an adaptive wavelength allocation pattern for NG-EPON. The impacts of different wavelength allocation patterns on network performances such as packet delay and bandwidth throughput are quantitatively studied. The rest of this study is organized as follows: Section 2 reviews some related work on dynamic wavelength and bandwidth allocation in NG-EPON. In Section 3, we introduce several wavelength allocation patterns for NG-EPON ONUs and analyze their respective strengthens and weaknesses. In Section 4, we quantitatively compare the performances of different wavelength allocation patterns. Finally, Section 5 concludes this study.
2. Related work
EPON technologies have evolved for many years. From 1G-EPON to 10G-EPON and hybrid WDM/TDM EPON, dynamic wavelength and bandwidth allocation for the upstream transmission operates in the similar request-and-grant way. In 1G-EPON and 10G-EPON which have only one wavelength, the problem of wavelength allocation doesn’t exist and only dynamic bandwidth allocation (DBA) needs to be arbitrated [6]. In hybrid WDM/TDM EPON, there are multiple wavelengths in the system. Each ONU is equipped with a tunable laser and can work on any available wavelength. ONUs can be allocated to work on some wavelength dynamically. The scheduling domains contain both timeslots as well as wavelengths [7]. Many algorithms are proposed to schedule ONUs in WDM/TDM EPONs, such as [8, 9].
It is noteworthy that NG-EPON architecture is very different from previous generations of EPONs. NG-EPON introduces multiple wavelengths and channel bonding to enable ONUs to transmit data on multiple wavelengths simultaneously. The feature of channel bonding makes NG-EPON ONUs different from previous ones and NG-EPON scheduling becomes more complicated. Although the NG-EPON scheduling is at early stages of research, the importance of well-designed DWBA algorithms for coordinating NG-EPON upstream transmission is undeniable since channel bonding is a brand new feature in NG-EPON. So far, there are several literatures on DWBA for NG-EPON. [10] investigates the requirements for NG-EPON bandwidth allocation. Literature [11] shows an offline DWBA algorithm for mitigating frame reordering problem. Besides, a DWBA algorithm for scheduling different kinds of NG-EPON ONUs is proposed in [12] and it only considers the bandwidth allocation between several kinds of ONUs. A DWBA algorithm called Water-Filling (WF) algorithm for NG-EPON is proposed in [13]. In the WF algorithm, wavelength resources are regarded as depression storages and ONU bandwidth requests are regarded as water. The dynamic wavelength and bandwidth allocation is similar to pouring water into connected depression storages. The WF DWBA algorithm can utilize wavelength resources efficiently, keep load balancing between wavelengths and achieve small packet delay. To reduce the occurrences of unnecessary guard time caused by the WF algorithm, a flexible wavelength allocation algorithm called FW algorithm is proposed in [14]. FW algorithm behaves similarly to WF algorithm except that the number of granted wavelengths to an ONU is flexible. If the allocated timeslot is smaller than the guard time size in one wavelength with WF-like allocation, then this wavelength will not be selected as granted wavelength. FW algorithm introduces flexible sets of wavelengths to ensure that the allocated bandwidth on each wavelength is larger than the guard time (or some fixed value). This can reduce unnecessary occurrences of guard time that decreases bandwidth resource utilization. Besides, considering abundant guard time introduced by WF algorithm, [15] proposes a First-Fit (FF) algorithm that allocates each ONU with the earliest available single wavelength. FF algorithm improves bandwidth utilization in some cases compared with WF algorithm. To the best of our knowledge, there are no researches about the comprehensive influence of wavelength allocation patterns for NG-EPON.
3. Wavelength allocation patterns in NG-EPON
In NG-EPON, 50G ONUs are equipped with two pairs of transceivers, which enable ONUs to transmit data on two wavelengths simultaneously. Since all the ONUs share the total available bandwidth, the OLT needs to arbitrate transmission wavelengths and timeslots for each ONU in the upstream direction. The OLT will inform an ONU of transmission wavelengths and timeslots by sending a GATE message to it. Although an ONU can receive GATE messages on all its capable wavelengths, only one GATE message is enough to carry information about granted transmission timeslots on each wavelength. After receiving GATE message from the OLT, the ONU will transmit data during assigned timeslots on each wavelength as GATE message indicates. At the end of its latest transmission timeslots, the ONU will send a REPORT message to the OLT to report its queued data size. DWBA in the OLT will calculate the transmission wavelengths and timeslots in next cycle for the ONU according to its request.
According to the scheduling trigger condition, DWBA can be divided into two major kinds: (a) offline DWBA that schedules ONUs only when all the ONUs’ bandwidth request are received and (b) online DWBA that schedules an ONU instantly when its bandwidth request is received. To avoid the service interrupt caused by collecting REPORT messages from all the ONUs, online DWBA is considered in this study. Therefore, when an ONU’s REPORT message arrives at the OLT, DWBA needs to calculate transmission wavelengths and timeslots for it immediately. We consider an NG-EPON that has W wavelengths and N ONUs. Each wavelength works at the rate of C. In this paper, we refer to 50G EPON, W = 2 and C = 25 Gb/s, although our discussion can be applied to more general conditions. To avoid DWBA cycle length L being too large, the total bandwidth assigned to an ONU should be limited. Assume the bandwidth limit for each ONU is LMTi, the bandwidth limit for each ONU should be:
According to the limited bandwidth allocation policy, an ONU i that requests for Ri will be assigned with bandwidth: After calculating total bandwidth Gi assigned to an ONU, DWBA needs to decide which wavelengths that the ONU will occupy to transmit its data. In this step, various wavelength allocation patterns can be adopted. Since 25G ONUs only work on single wavelength and there is no wavelength allocation problem for them, we focus on 50G ONUs that have the flexibility of working on single or multiple wavelengths.3.1. Multiple wavelength allocation
In the multiple wavelength allocation, each ONU will be assigned with more than one wavelength each time. Each ONU’s timeslots will be distributed among multiple wavelengths. There can be many ways to distribute an ONU’s timeslot among multiple wavelengths, and among them Water Filling (WF) [13] is the most well-known way. To achieve the smallest packet delay, each allocation tries to make every bit transmit as earlier as possible. As Fig. 2(a) shows, ONU #1, #2, #3 and #4 are allocated with all the two wavelengths λ1 and λ2. Such allocation eliminates occupation differences of all wavelengths and all the ONUs will be allocated with two wavelengths finally. Intuitively, WF wavelength allocation pattern can utilize the equipped lasers to achieve higher throughput. However, since guard time between two consequential timeslots is essential, fragmentation of timeslots can introduce bandwidth waste due to too much guard time. FW algorithm [14], which can be regarded as an improved version of WF algorithm, behaves similarly to WF algorithm except the number of granted wavelengths is flexible for the sake of bandwidth utilization.
Fig. 2 Wavelength allocation for ONUs (marked as 1, 2, 3 and 4) with different wavelength allocation patterns.
3.2. Single wavelength allocation
In the single wavelength allocation, each ONU will be assigned with only one wavelength. Although the 50G ONUs are equipped with two transceivers, they are allocated with single wavelength for data transmission, which is very similar to that in TWDM-PON [16]. To achieve smaller packet delay, the earliest available wavelength will be assigned to each ONU, which is used in the First-Fit (FF) algorithm in [15]. Figure 2(b) shows an example of FF wavelength allocation pattern. All the ONUs are allocated with their earliest available wavelengths. FF wavelength allocation pattern is easy to implement since there is no frame demultiplexing and multiplexing among multiple wavelengths. Besides, FF reduces the occurrences of guard time as Fig. 2(b) shows. However, FF limits ONUs’ peak rates because ONUs are always allocated with only one wavelength.
3.3. Hybrid wavelength allocation
Considering single and multiple wavelength allocation patterns have their respective strengths and weaknesses, it is reasonable to combine them dynamically. To this end, we introduce a threshold-based dynamic single and multiple wavelengths allocation pattern. When an ONU’s bandwidth request arrives, the OLT will allocate it with certain amount of bandwidth as Eq. (2) illustrates. In the following wavelength allocation, the OLT will compare the granted bandwidth size Gi with a threshold τ to decide either single wavelength allocation or multiple wavelengths allocation. Since ONU’s grant size Gi is within bandwidth limit LMTi, we use a ratio γ (0 ≤ γ ≤ 1) to represent the threshold for wavelength allocation.
By comparing ONU’s bandwidth grant size and wavelength allocation threshold, the OLT will make corresponding wavelength allocation decisions: If an ONU’s grant size is larger than a threshold τ, the ONU will be assigned with multiple wavelengths. Otherwise, the ONU will be assigned with single wavelength. As Fig. 2(c) shows, the grant sizes of ONU #1 and ONU #3 are smaller than threshold (G1 < τ and G3 < τ) while the grant sizes of ONU #2 and #4 are larger than threshold (G2 > τ and G4 > τ). Therefore, ONU #1 and #3 are allocated with single wavelength while ONU #2 and #4 are allocated with multiple wavelengths. On one hand, by allocating grants that are smaller than the threshold with single wavelength, hybrid wavelength allocation reduces the occurrences of guard time due to scattering of each ONU’s timeslot on multiple wavelengths. On the other hand, hybrid wavelength allocation pattern frees those ONU grants that are larger than the threshold to engage multiple wavelengths for higher peak rates.It is noteworthy that γ in hybrid wavelength allocation pattern is very important. As Fig. 3 shows, γ in hybrid wavelength allocation pattern can be set to any value between 0 and 1 arbitrarily. When γ = 0, all the ONUs will always be allocated with multiple wavelengths, which is WF. When γ = 1, all the ONUs will always be allocated with single wavelengths, which is FF. From this aspect, WF and FF are special cases (γ = 0, 1 respectively) of hybrid wavelength allocation pattern. For other values of γ, a combination of both single wavelength allocation and multiple wavelength allocation will be implemented. If γ is closer to 0, ONUs are more likely allocated with multiple wavelengths. On the contrary, if γ is closer to 1, ONUs are more likely allocated with single wavelength. For hybrid wavelength allocation pattern, the main objective is to find the appropriate threshold ratio γ for the best network performance.
Fig. 3 Hybrid wavelength allocation pattern (0 ≤ γ ≤ 1) can unify WF (γ = 0) and FF (γ = 1) wavelength allocation patterns.
3.4. Adaptive wavelength allocation
Hybrid wavelength allocation pattern uses static threshold ratio γ to judge whether an ONU will be allocated with single wavelengths or multiple wavelengths. Considering the fluctuation and tidal effect of user traffic, a static threshold ratio γ cannot be perfectly applied to all scenarios. Therefore, we propose a wavelength allocation pattern that dynamically allocates ONUs with single or multiple wavelengths based on an adaptive threshold.
For an NG-EPON with N ONUs, the OLT maintains a N dimensional array to record and update the bandwidth allocated to ONUs each time. After determining the allocated bandwidth G i for a newly-arrived ONU’s request, the OLT will determine the working wavelengths for this ONU. First, update the allocated bandwidth of ONU i and calculate the sum of allocated bandwidth of all ONUs.
Then calculate the threshold ratio γ as follows:It is worth mentioning that all the grants except G i have been allocated with wavelengths. Then compare the threshold with the allocated bandwidth of the newly-arrived ONU. If Gi ≤ γ ∗ LMTi, then allocate the newly-arrived ONU (e.g. ONU #4 in Fig. 4) with single wavelength. Otherwise, allocate the newly-arrived ONU (e.g. ONU #1 in Fig. 4) with multiple wavelengths. In each wavelength allocation, the threshold ratio γ for determining working wavelengths is recalculated. The reason for choosing adaptive threshold ratio in Eq. (5) is that adaptive wavelength allocation pattern tries to balance load among all the W wavelengths. The adaptive wavelength allocation takes into account both the absolute value of an ONU’s request and the relative value to other ONU requests.
Fig. 4 Adaptive wavelength allocation pattern: (a) single wavelength allocation if Gi ≤ Sum/W ; (b) multiple wavelength allocation if Gi > Sum/W.
The key problem that wavelength allocation patterns deal with is how to allocate ONU’s bandwidth grant on the wavelengths. Different wavelength allocation patterns can decide single wavelength or multiple wavelengths for an ONU depending on their respective logics. The detailed wavelength and bandwidth allocation algorithm is illustrated in Algorithm 1. First, DWBA will calculate Gi and corresponding γ. Then adjust scheduled time for ONU i by considering round-trip time. After sorting scheduled time on each wavelength as , DWBA will judge whether single wavelength or multiple allocation will be implemented. IfTGi ≤ γ ∗ LMTi, then allocate this ONU with the earliest wavelength in the FF-like way. Otherwise, allocate this ONU with multiple wavelength in the WF-like way. And its key principle is trying to eliminate the differences of start times on different wavelengths and make every bit transmitted as earlier as possible. The algorithm fills ONU’s granted bandwidth to the earliest available wavelength to make its start time the same as the second earliest available one. Then fill the earliest two wavelengths to make their start time the same as the third one. Repeat these steps until the start time difference of all the wavelengths has been eliminated or the remaining bandwidth is not enough to equal remaining wavelength start time. Then allocate remaining bandwidth equally to all the equaled wavelengths.
Algorithm 1. Algorithm for wavelength allocation
As Fig. 4(b) shows, the newly-arrived ONU #1 will be allocated with multiple wavelengths. First, allocate ONU #1 with 150 KB on wavelength λ1 to eliminate the bandwidth occupation differences of λ1 and λ2. Then, distribute the remaining granted bandwidth (700-150=550 KB) of ONU #1 equally among two wavelengths. So the final granted bandwidth on λ1 and λ2 is 425KB and 275KB respectively. It is notable that multiple wavelength allocation doesn’t always allocate all the W wavelengths to the ONU. The number of allocated wavelengths depends on the detailed bandwidth occupation on each wavelength and ONU grant size. In either single or multiple wavelength allocation, wavelength and bandwidth decisions for each ONU are made out of transmitting its data as earlier as possible and minimizing of its packet delay. The output of Algorithm 1 is the wavelength and bandwidth allocation results for ONU i. To be specific, BW[w][0] stands for transmission start time on wavelength w, and BW [w][1] stands for granted bandwidth on wavelength w. The wavelength allocation algorithm can be implemented with O(W) time, in which W represents the wavelength number.
4. Performance evaluation of different wavelength allocation patterns
Different wavelength allocation patterns can have different impacts on network performances. It is very important to figure out which wavelength allocation patterns are most suitable for NG-EPON. Simulations are conducted to quantitatively evaluate the impacts of different wavelength allocation patterns on network performances. A NG-EPON system with two wavelengths serving 16 ONUs is simulated. In order to be more rigorous and realistic, bursty traffic is considered in our simulations. Each ONU generates self-similar traffic [17] with a Hurst parameter of 0.75 and the packet size follows the same distribution in [6]: 60% 64 bytes, 4% 300 bytes, 11% 580 bytes and 25% 1518 bytes. Each ONU has a buffer with the size of 20 Mbit. The Round-Trip Times (RTTs) for ONUs are uniformly distributed between 0 and 200 µs, which represents that the max OLT-ONU distance is around 20 kilometers. The guard time that avoids overlapping of two sequential timeslots of different ONUs is set to be 5 µs. The bandwidth limit LMT i for each ONU is 6250000 bit during each DWBA cycle. We quantitatively compare network performances (packet delay, throughput and packet loss ratio) with both all the mentioned wavelength allocation patterns. For hybrid wavelength allocation patterns, different threshold ratios γ (0, 0.01, 0.1 and 1) are simulated. Specially, γ = 0 and γ = 1 stand for WF and FF pattern respectively. Proposed adaptive wavelength allocation pattern is maked as “Adaptive γ” in the legend in the following result figures. Besides, the FW wavelength allocation pattern, marked as “FW”, is also compared. Considering the ONU number, the alpha factor in FW algorithm [14] is set as 1. First, we consider the scenario in which only 50G ONUs exist in the PON. Then, we discuss the coexistence scenario in which both 25G and 50G ONUs exist in the same PON.
4.1. Pure 50G ONUs scenario
4.1.1. Uniform load cases
First, we compare different wavelength allocation patterns in uniform load cases, in which ONUs generate similar amount of traffic. As shown in Fig. 5(a), different threshold ratios γ can result in different packet delay. When γ = 0 (i.e., WF), ONUs experience the largest packet delay. When γ = 1 (i.e., FF), ONUs experience the smallest packet delay. When γ is between 0 and 1, the packet delay varies between that of γ = 0 and γ = 1. As an improved version of WF, FW has smaller packet delay when the total load is low. However, when the total load becomes high, FW has the same delay performance as WF.
Fig. 5 Network performance of 50G ONUs with different wavelength allocation patterns in uniform load cases.
Generally, a larger γ will lead to a smaller packet delay. It is easy to explain why transmitting on multiple wavelengths increases packet delay. In the WF pattern, each timeslot will be broken into pieces and distributed on multiple wavelengths. Such allocation will introduce abundant guard time, which increases total time that an ONU waits for the next data transmission. On the contrary, less guard time will be needed if ONU timesolts are allocated on single wavelengths. Same to FF, adaptive γ has the smallest packet delay. In uniform load cases, ONUs’ requests are similar to each other. As a consequence, adaptive wavelength allocation pattern works in the same manner as FF. Despite different wavelength allocation patterns cause different packet delay, the total throughput for different wavelength allocation patterns shown in Fig. 5(b) is close to each other, which represents similar bandwidth utilization for different wavelength allocation patterns. When the total load is very high (>40 Gb/s), ONUs begin to drop packets. As Fig. 5(c) shows, FF and adaptive wavelength allocation pattern has smaller packet drop ratio than others. This is because they provide more available bandwidth for ONUs by reducing certain amount of guard time.
4.1.2. Non-uniform load cases
In most cases, some heavy users, such as large enterprises and companies that need massive data transmission, can have huge bandwidth demands. It is reasonable to consider that some ONUs have much higher load than the average level. Therefore, we also evaluate network performances of different wavelength allocation patterns in non-uniform load cases, in which an ONU has pretty high load. The background load is set as around 10 Gb/s and a target ONU’s load will increase gradually.
As shown in Fig. 6(a), the packet delay of the target ONU varies with different wavelength allocation patterns. For γ = 1 (i.e., FF), the packet delay is the smallest when the target ONU’s load is under 5 Gb/s. However, when the target ONU’s load increases to 10 Gb/s, packet delay with FF pattern increases rapidly. Due to buffer limit, packet delay with FF becomes stable (but not acceptable) when the load is larger than 30 Gb/s. γ = 0.1, adaptive γ and FW always have the smallest packet delay. The other threshold ratios lead to a larger packet delay. Corresponding throughput of target ONU is shown in Fig. 6(b). When γ = 1 (i.e., FF), target ONU can only achieve around 20 Gb/s maximum throughput, even if it offers much higher load. For other γ, target ONU can achieve much higher bandwidth throughput. In general, FF can not provide ONUs with more than 16 Gb/s throughput, while other wavelength allocation patterns can provide much higher throughput (around 25 Gb/s). It is because FF pattern limits each ONU to work on only one wavelength each time. Since each wavelength works at the rate of 25 Gb/s, no ONU working on single wavelengths can achieve throughput higher than 25 Gb/s even if this ONU occupies a wavelength all the time. However, for other wavelength allocation patterns, an ONU with huge bandwidth demand can utilize multiple wavelengths to transmit its data, which helps the ONU to achieve higher peak rates. Since the FF pattern has limited throughput, the ONU has to drop more packets due to buffer limit with FF pattern, which results in much higher packet loss ratio as illustrated in Fig. 6(c).
Fig. 6 Network performance of target 50G ONU with different wavelength allocation patterns in non-uniform load cases.
In addition, the impacts on other ONUs are also studied. Figure 7(a) shows average packet delay of other ONUs in non-uniform load cases. γ = 1 (i.e., FF) has the smallest packet delay since the heavy-loaded target ONU occupies only one wavelength and other ONUs suffer less preemption. On the contrary, γ = 0 and γ = 0.01 have the largest packet delay since target ONU preempts certain amount of bandwidth on every wavelength. The packet delay with other γ varies between that of γ = 0 (i.e., WF) and γ = 1 (i.e., FF). The different impacts of wavelength allocation patterns on other ONUs’ delay can be explained from two aspects. On one hand, the smaller γ is, the more guard time will be introduced, which makes ONUs to wait for longer time for data transmission. On the other hand, smaller γ leads to more ONUs transmitting data on multiple wavelengths. With the FF pattern, the heavy-loaded target ONU only transmits data on single wavelength. However, with the other patterns, the heavy-loaded ONU transmits data on all the wavelengths which will delay other ONUs’ traffic and cause larger packet delay of other ONUs. Nevertheless, the degradation of other ONUs’ delay performance [0.1 ms in Fig. 7(a)] is insignificant when compared with the improvement of target ONU’s delay performance [0.5 ms in Fig. 6(a)]. Figure 7(b) shows the average throughput of other ONUs in non-uniform load cases. Generally, all the γ result in similar throughput. Since these ONUs’ load is not very high and their bandwidth demands are always satisfied, no packet will be dropped.
Fig. 7 Network performance of other 50G ONUs with different wavelength allocation patterns (represented by different γ) in non-uniform load cases.
4.1.3. Impacts of guard time size
To avoid two sequent timeslots overlapping with each other, guard time is needed. As analyzed previously, WF (γ = 0) and FF (γ = 1) are the two extremes among all the γ, in both uniform and non-uniform load cases. With WF patterns, large amount of guard time is needed because each timeslot of an ONU is broken into pieces and distributed on multiple wavelengths. While with FF pattern, less guard time are needed since each ONU are allocated with one entire timeslot on single wavelengths. Therefore, the impacts of guard time on different allocation patterns need to be studied. We compare the packet delay with different guard time sizes (namely 1 µs, 5 µs and 10 µs) and two typical wavelength allocation patterns (WF and FF). As shown in Fig. 8, larger guard time results in larger packet delay in both WF and FF allocation patterns. The delay difference caused by different guard time sizes in WF pattern is more significant than that in FF pattern. This is because WF pattern introduces much more guard time.
Fig. 8 Average packet delay versus guard time size with two extreme wavelength allocation patterns: WF (γ = 0) and FF (γ = 1).
4.2. 25G and 50G ONUs coexistence scenario
Since 25G ONUs can coexist with 50G ONUs in the same PON, we also evaluate the performances of different wavelength allocation patterns in the 25G and 50G ONUs coexistence scenario. Because 25G ONUs work on single wavelength, there is no wavelength allocation problem for 25G ONUs. We study the impacts of different wavelength allocation patterns of 50G ONUs on the single-wavelength 25G ONUs. In our simulations, there are eight 25G ONUs and eight 50G ONUs. 25G ONUs offer total 10Gb/s traffic load. We evaluate the network performances of 25G ONUs while 50G ONUs employ different traffic load and wavelength allocation patterns.
As Fig. 9(a) shows, the packet delay of 25G ONUs increases with 50G ONUs’ traffic load. This is because when the 50G ONUs have higher load (>25 Gb/s), more bandwidth on wavelength λ1 will be shared or deprived by 50G ONUs. However, among all the wavelength allocation patterns, FF (γ = 1) and adaptive (Adaptive γ) wavelength allocation patterns show the minimal impact on 25G ONUs’ packet delay. FF and adaptive wavelength allocation patterns reduce the occurrences of guard time, and more bandwidth could be utilized for data transmission. As Fig. 9(b) shows, the throughput of 25G ONUs with different 50G ONUs’ wavelength allocation patterns is similar to each other. The bandwidth demand of 25G ONUs are satisfied and no packet of 25G ONUs gets dropped as Fig. 9(c) shows.
Fig. 9 Network performances of 25G ONUs when 50G ONUs employ different wavelength allocation patterns in the coexistence scenario.
Besides, the 50G ONUs’ network performances in the coexistence scenario are also studied. As shown in Fig. 10(a), γ = 1 and Adaptive γ wavelength allocation patterns have the smallest packet delay, which is similar to that in the pure 50G ONUs scenario. The throughput of 50G ONUs shown in Fig. 10(b) is close to each other. When the total load reaches 40 Gb/s, the γ = 1 and Adaptive γ wavelength allocation patterns have larger throughput than other wavelength allocation patterns. Accordingly, as Fig. 10(c) shows, the packet loss ratio of these two wavelength allocation patterns is smaller than that of others.
Fig. 10 Network performances of 50G ONUs when 50G ONUs employ different wavelength allocation patterns in the coexistence scenario.
Generally, the performances of different wavelength allocation patterns in 25G and 50G ONUs coexistence scenario has similar trend compared with that in pure 50G ONUs scenario. Different γ values lead to different network performances. γ = 0 has better performance in non-uniform load cases while γ = 1 has better performance in uniform load cases. The performance of other γ values is between that of the two extreme γ values. FW wavelength allocation pattern can improve network performance compared with WF (γ = 0) when the load is low. Among all the wavelength allocation patterns, only proposed adaptive wavelength allocation pattern can achieve the optimal network performances including packet delay, throughput and packet loss ratio.
5. Conclusion
In this study, we investigated online upstream dynamic wavelength and bandwidth allocation for NG-EPON. We used a threshold-based hybrid wavelength allocation to unify the existing wavelength allocation patterns (WF and FF). Besides, we proposed an adaptive wavelength allocation pattern that can allocate ONUs dynamically according to an adaptive threshold. Simulations are conducted in both pure 50G ONUs scenario and 25G and 50G ONUs coexistence scenario. Simulation results show that our proposed adaptive wavelength allocation pattern outperforms other wavelength allocation patterns. Specifically, adaptive wavelength allocation pattern achieves small packet delay, large bandwidth throughput and low packet loss ratio. We also investigate the impacts of guard time size on network performances. Generally, larger guard time leads to larger ONU packet delay, especially for the multi-wavelength allocation.
Funding
National Nature Science Foundation of China (NSFC) (61471238, 61431009, 61221001, 61371082).
Disclosures
The authors declare that there are no conflicts of interest related to this article.
References
1. IEEE P802.3ca 100G-EPON Task Force, “Physical layer specifications and management parameters for 25 Gb/s, 50 Gb/s, and 100 Gb/s passive optical networks,” http://www.ieee802.org/3/ca/.
2. F. J. Effenberger, “Industrial trends and roadmap of access,” J. Lightwave Technol. 35, 1142–1146 (2017).
3. “IEEE standard for information technology–local and metropolitan area networks–Part 3: CSMA/CD access method and physical layer specifications amendment: Media access control parameters, physical layers, and management parameters for subscriber access networks,” IEEE Std 802.3ah-2004 pp. 1–640 (2004).
4. “IEEE standard for information technology–local and metropolitan area networks–specific requirements–Part 3: CSMA/CD access method and physical layer specifications amendment 1: Physical layer specifications and management parameters for 10 Gb/s passive optical networks,” IEEE Std 802.3av-2009 (Amendment to IEEE Std 802.3-2008) pp. 1–227 (2009).
5. G. Kramer, “Frame latency issues in multilane EPON,” http://www.ieee802.org/3/ca/public/meeting_archive/2016/01/kramer_3ca_1a_0116.pdf.
6. M. P. McGarry and M. Reisslein, “Investigation of the DBA algorithm design space for EPONs,” J. Lightwave Technol. 30, 2271–2280 (2012).
7. A. R. Dhaini, C. M. Assi, M. Maier, and A. Shami, “Dynamic wavelength and bandwidth allocation in hybrid TDM/WDM EPON networks,” J. Lightwave Technol. 25, 277–286 (2007).
8. M. P. McGarry, M. Reisslein, C. J. Colbourn, M. Maier, F. Aurzada, and M. Scheutzow, “Just-in-time scheduling for multichannel EPONs,” J. Lightwave Technol. 26, 1204–1216 (2008).
9. K. Kanonakis and I. Tomkos, “Online upstream scheduling and wavelength assignment algorithms for WDM-EPON networks,” in 2009 35th European Conference on Optical Communication, (2009), pp. 1–2.
10. G. Krammer, “Flexible and extensible architecture for multiple generations of NG-EPON,” http://www.ieee802.org/3/ca/public/meeting_archive/2016/03/kramer_3ca_1_0316.pdf.
11. W. Wang, W. Guo, and W. Hu, “Dynamic wavelength and bandwidth allocation algorithms for mitigating frame reordering in NG-EPON,” J. Opt. Commun. Netw. 10, 220–228 (2018).
12. W. Wang, W. Guo, and W. Hu, “A fair and flexible dynamic wavelength and bandwidth allocation algorithm for IEEE 100G-EPON,” in 2017 19th International Conference on Transparent Optical Networks (ICTON), (2017), pp. 1–4.
13. L. Wang, X. Wang, B. Mukherjee, H. S. Chung, H. H. Lee, and S. Park, “On the performance of Hybrid-PON scheduling strategies for NG-EPON,” in 2016 International Conference on Optical Network Design and Modeling (ONDM), (IEEE, 2016), pp. 1–5.
14. S. B. Hussain, W. Hu, H. Xin, A. M. Mikaeil, and A. Sultan, “Flexible wavelength and dynamic bandwidth allocation for NG-EPONs,” J. Opt. Commun. Netw. 10, 643–652 (2018).
15. S. B. Hussain, W. Hu, H. Xin, and A. M. Mikaeil, “Low-latency dynamic wavelength and bandwidth allocation algorithm for NG-EPON,” J. Opt. Commun. Netw. 9, 1108–1115 (2017).
16. Y. Luo, X. Zhou, F. Effenberger, X. Yan, G. Peng, Y. Qian, and Y. Ma, “Time- and wavelength-division multiplexed passive optical network (TWDM-PON) for next-generation PON stage 2 (NG-PON2),” J. Lightwave Technol. 31, 587–593 (2013).
17. K. Park and W. Willinger, Self-Similar Network Traffic and Performance Evaluation (John Wiley & Sons, Inc., 2000).
