Open Access
Add to BibSonomyAbstract
To evaluate the cost efficiency of IP over elastic optical network architectures, we use a multi-layer network design scheme that covers network to node equipment level. An evaluation in a static traffic environment shows that the multi-flow optical transponder-based elastic optical network reduces total cost as well as equipment counts compared to other elastic network models based on fixed-rate, mixed-line-rate and bandwidth-variable transponders.
© 2013 Optical Society of America
1. Introduction
The need for cost- and energy-efficient networks continues to increase with the recent explosion in traffic demands. Using optical paths to offload IP traffic to the optical layer, which reduces the required number or capacity of costly and power-hungry core routers, has been widely studied as a way to construct economic backbone networks. However, the use of such paths in traditional fixed-rate optical networks is permitted only when there is sufficient traffic between routers to fill the entire capacity of an optical path. One way to extend the use of optical paths is to employ different types of fixed-rate transponders that allow the introduction of multi-granular optical paths, which is referred to as the mixed-line-rate approach [1]. Fixed-rate transponders are mature and costs are low, but there is a concern regarding this approach in terms of inventory management given the different types of and large numbers of transponders. Another approach is to assign the minimum spectral width necessary to support each optical path demand by bandwidth-variable transponders (BVTs) [2]. This network architecture is becoming feasible with the standardization of the flexible grid concept in ITU-T, and recent technological progresses in realizing optical switching at arbitrary bandwidth granularity. This architecture aims to maximize spectral efficiency, but its unified transponder architecture may lack transponder utilization efficiency because fails to consider the capacity of optical path demands; for example, a 400 Gb/s capacity BVT may accommodate only 40 Gb/s optical path demand. A recently proposed multi-flow transponder (MFT) [3] is expected to mitigate these concerns because it can fully utilize the transponder resource by generating a wide variety and multiple optical flows in a single transponder. This is realized by slicing parallelized sub-transceivers into virtual sub-transponder resources. The effectiveness of this transponder has been studied [3–5].
Our previous work reported the impact of MFT in terms of the necessary number of sets of equipment such as router interfaces, transponders, and wavelength selective switches (WSSs) [5]. However, the multi-layer network design scheme employed in the work was optimized for router interface count, not total cost. In this paper, we compare four IP over elastic optical network architectures by a cost optimization scheme, and show the benefit of elastic optical networks based on MFTs in terms of cost as well as equipment count. The rest of this paper is organized as follows. Section 2 describes the equipment models. Section 3 introduces the multi-layer network design scheme. Section 4 gives the evaluation settings and results. Finally, Section 5 concludes this paper.
2. Transponder, (BV-)WXC, and IP router models
Figure 1 shows the multi-layer equipment structure considered here; it consists of wavelength cross-connects (WXCs), transponders, and IP routers. We compare four elastic network models characterized by the transponder type employed: the Fixed-Rate (Fixed) model, Mixed-Line-Rate (MLR) model, Bandwidth-Variable (BV) model, and Multi-flow (MF) model. The Fixed model uses 40, 100, or 400 Gb/s fixed-rate transponders to support a wide variety of traffic demands. The MLR model supports 40, 100, and 400 Gb/s fixed-rate transponders, and selects the most appropriate transponder according to the optical path demand. The BV model accommodates all types of optical path demands with 100 or 400 Gb/s BVTs in a spectrally efficient manner. As for the MF model, a MFT can make optical connections with different nodes, which is realized by a flow distributor and multiple parallelized sub-transceivers [3]. The flow distributor switches incoming traffic from an IP router into multiple data flows, each of which has a different destination, and the necessary number of sub-transceivers is assigned to each data flow to generate a multi-carrier optical channel, or optical flow, based on OFDM or Nyquist-WDM. Employing these technologies is now seen as a practical approach for beyond-100 Gb/s transponders [3]. Changing the number of active sub-carriers enables the rate-adaptive function of BVT and MFT.
The necessity of bandwidth flexibility in the optical node differs with the network model. As the fixed model does not require bandwidth flexibility, the model employs conventional WXC. Meanwhile, the three other models use a bandwidth-variable WXC (BV-WXC) to support various optical path capacities. A (BV-)WXC is assumed to consist of a cross-connect block and one or more add/drop block(s). The (BV-)WXC has colorless and directionless capabilities [6], and avoids inner-node wavelength contention by implementing additional add/drop blocks before it occurs. This work assumes that each location has only one (BV-)WXC, which can have an arbitrary number of add/drop blocks and output fibers. Two (BV-)WSSs are equipped at an add/drop block. This work assumes that each (BV-)WSS in an add/drop block has the port count of 1 x 8; this value is the result of a tradeoff because lower port counts require many (BV-)WSSs, while larger port counts increase the frequency of inter-node wavelength blocking. Note that many available WSSs have 1 x 9 ports, but this work matches (BV-)WSS port count to a practical optical coupler arrangement (1 x 8). In the cross-connect block, (BV-)WSS port count is flexible and is determined by the result of optical path allocation.
We assume that an IP router simply consists of a chassis and one or more line card(s); other modules such as switching fabric and routing engine are assumed to be mounted in the router chassis. An IP router line card contains several interfaces that connect to transponders or clients. This work examines three types of line cards, each of which has the same capacity: 40 Gb/s x 10 ports, 100 Gb/s x 4 ports, and 400 Gb/s x 1 port. Each line card accommodates the same type of transponders. The MLR model selects transponders based on the capacity of optical path demands. On the other hand, the Fixed, BV, and MF models only use single type of line card according to given transponder capacity.
3. Multi-layer network design scheme
The heuristic scheme described in this section is used to optimize the IP and optical multi-layer network laid over the given physical fiber link topology. This scheme yields a nearly optimal IP layer (logical layer) and optical layer (physical layer) topology in terms of minimized total costs given static traffic demands. The scheme is based on the composition method [7], which adds IP layer connection from the minimum IP layer topology, assigns optical paths, and calculates a metric iteratively. The steps are given hereafter.
- 1. First, construct the minimum spanning tree in the IP layer such that any node can reach any other node.
- 2. Assign traffic demands along the IP layer topology and generate optical path demands by summing all traffic demands in each IP layer link. The traffic demands are assigned along the shortest route. If an optical path demand exceeds the maximum transponder capacity, it is divided into multiple optical path demands.
- 3. Determine appropriate routes and frequency slots from the optical path demands by the routing and spectrum assignment (RSA) algorithm on the physical fiber link topology. This work employs the first-fit algorithm for simplicity. The RSA algorithm also determines which add/drop blocks are to be used at both ends of the (BV-)WXCs. Since these add/drop blocks do not have contentionless capability in this study, frequency slots cannot be overlapped in each of them. An add/drop block is added when all (BV-)WSS ports in all add/drop blocks are utilized or when the frequency slots to be assigned are not available for all add/drop blocks. If we have the option of expanding fiber number or add/drop blocks to accommodate an optical path demand, the scheme prioritizes the latter approach, for the former is usually more costly than the latter.
Moreover, in the MF model, the scheme determines which transponder the optical path demand should be assigned to, because a used MFT can support additional optical flows if it has residual transponder capacity. This work selects the MFT that has the smallest difference between the residual capacity and the optical path demand. Only if there is no transponder that can accommodate the optical path demand is an additional transponder introduced.
- 4. When route and spectrum slots for each optical path are determined, the number of sets of equipment required to accommodate these traffic demands is calculated. After determining the required number of equipment sets, total cost is calculated and recorded.
- 5. Add an extra IP layer connection to the IP layer topology and return to step 2, until any pair of IP routers is directly connected. The order of adding IP layer connection is determined from the product of hop counts and traffic volume between nodes in descending order.
These steps calculate the total costs for the entire multi-layer topology. The topology with lowest total cost is taken to be optimal.
4. Evaluation settings and results
This work evaluated the cost and necessary number of equipment sets determined by the four network models for the COST-266 Pan-European topology. Each fiber has 4 THz capacity. Non-uniform base traffic distribution between each node pair is derived based on reference [8]. The average traffic demand between nodes varies from 5 Gb/s to 150 Gb/s and is determined by multiplying the base traffic by different factors. This study assumes that all optical paths are unprotected and can be connected without regeneration for simplicity. Introduction of distance-adaptive modulation [9] is a future work.
The frequency slot width for the given optical path demands in each model, see Table 1, and the equipment cost, shown in Table 2, are mainly based on [3] and [10], respectively. The Fixed and BV model use a single type of transponder for each evaluation. The former consumes uniform frequency slot width regardless of the given optical path demands, and the latter assigns the minimum amount of frequency slot width sufficient for the optical path demand. The MLR model selects different kinds of transponders so as to minimize the total cost of accommodating the optical path demand. In the MF model, 1 MFT contains 10 sub-transceivers, each with capacity of 40 Gb/s. It can generate various optical flows by slicing the sub-transceivers to yield combinations such as 400 Gb/s x 1 flow, 240 Gb/s + 160 Gb/s, or 40 Gb/s x 10 flows according to optical path demand. MFT cost is currently not available, so this work assumes it to be 1.0 – 1.5 times 400 G BVT cost. The IP router can be extended from the single-chassis to the multi-chassis configuration, and we refer to the general formula described in [10] for calculating IP router chassis cost. For simplicity, (BV-)WXC cost is assumed to be determined by the sum of component (BV-)WSS costs. Only Fixed model employs WSSs, the other models BV-WSSs.
Table 1. Assigned frequency slot width (GHz) for optical path demand (Gb/s)
Table 2. Equipment cost
4.1 Preliminary evaluation
Prior to comparing the four network models, appropriate transponder specs for the Fixed and BV models are determined to ensure a fair comparison. We assumed 40 / 100 / 400 G transponder capacity for the Fixed model, and 100 / 400 G for the BV model.
Figure 2 shows the cost comparison of Fixed models. We can see from Fig. 2(a) that Fixed 40 G model is slightly superior to other models when traffic intensity is small, but that Fixed 400 G model outperforms other models when traffic is heavy. Figure 2 (b) explains the cost breakdowns when average inter-node traffic demand is 100 Gb/s. The Fixed 400 G model requires more router resources than the other two models because a single router line card cannot accommodate multiple 400 G transponders, but it reduces transponder and WXC costs. The savings provided by the Fixed 40 G and 100 G models in terms of router line card cost could not fully offset the increase of transponder cost. Similar results were obtained for the BV models (Fig. 3).
Fig. 2 Total cost comparison of three Fixed models. (a) Cost as a function of average inter-node traffic demand. (b) Cost breakdowns when average inter-node traffic demand is 100 Gb/s.
Fig. 3 Total cost comparison of two BV models. (a) Cost as a function of average inter-node traffic demand. (b) Cost breakdowns when average inter-node traffic demand is 100 Gb/s.
Based on these results, we can adopt the 400 G transponder for both Fixed and BV model in the following subsection.
4.2 Evaluation among Fixed, MLR, BV, and MF models
Figure 4 shows the total cost (i.e. sum of all equipment cost) as a function of the number of IP layer connections added to the initial spanning tree when the average inter-node traffic demand is 100 Gb/s. Optimal points (i.e. minimum cost) for the network models are also shown. Increasing the number of added IP layer connections means that intermediate router processing in IP layer connections is offloaded and the number of optical paths directly connected between nodes is increased. As IP layer connections are added to the initial state, all models offer lower cost, until the optimal point is reached, which differs among the models. The optimal IP layer topologies for each model are illustrated in Fig. 5. They correspond to the optimal points of Fig. 4. The Fixed and BV model have almost the same topology because they have no way of accommodating traffic efficiently but instead groom it in the IP layer. Therefore, the two models try to maximize the capacity of IP layer connections and reduce the number of line cards and transponders. On the other hand, the MF models yield close to full mesh topologies if the capacity of each optical path is low, which explains why MFTs create a lot of low-rate optical connections. The MLR model lies between these two characteristics. The model can select the appropriate type of transponder according to optical path demand, but its granularity is lower than the number of optical flows that MFT can generate.
Fig. 4 Total cost as a function of the number of IP layer connections added to the initial minimum spanning tree, and optimal point for each model. Average inter-node traffic demand is 100 Gb/s. MFT cost of illustrated MF model is 33.
Fig. 5 IP layer topologies for each optimal point illustrated in Fig. 4. (a) Fixed model. (b) MLR model. (c) BV model. (d) MF model.
Figure 6(a) shows the total cost in the optimum topology as a function of the average inter-node traffic demand. In the MF model, two graphs are plotted; MFT cost in the upper bound and lower bound is 33 and 22, respectively. We can see that the MF model saves cost compared to the three other models for all traffic intensities regardless of MFT cost. We should note that the BV model is more expensive than the Fixed model is because BVT is more expensive than fixed-rate transponders, and fiber installation cost is not considered despite BVT’s higher spectral efficiency. The aspect of spectral efficiency will be discussed later in detail. Cost breakdowns are illustrated in Fig. 6(b). Router card costs are greatly reduced in the MF model.
Fig. 6 Total cost comparison. (a) Cost as a function of average inter-node traffic demand. (b) Cost breakdowns when average inter-node traffic demand is 100 Gb/s. MFT cost of illustrated MF model is 33.
Figure 7 shows the percentages of necessary number of equipment sets with reference to the Fixed model when the inter-node average traffic demand is 100 Gb/s. The MLR model requires more transponders, BV-WSSs, and fibers than the three other models, which is natural because the model employs a lot of low-rate transponders to connect nodes without grooming at the IP layer as shown in Fig. 5(b). The increase use of optical paths in the MLR model yields an increase in BV-WSSs and fibers, but it does not impact total cost so much because this work assumes that a router line card, which is the dominant factor determining total cost, can support more than one low rate transponder in the MLR model.
Fig. 7 Necessary equipment counts relative to Fixed model. Average inter-node traffic demand is 100 Gb/s.
The BV model outperforms the three other models in terms of the number of fibers due to its spectrum-efficient optical path allocation. We can see its effect from the number of total allocated frequency slot units [11] in Fig. 8. This work calculates this metric by multiplying frequency slot units by path hops and summing the values up for all paths. The reason why the BV model saves the number of fiber compared to the MF model even though transponder-level elasticity in the BV model is the same as the MF model as shown in Table 1 is that the BV model creates larger capacity optical paths, which leads to higher spectral efficiency, than the MF model. However, the difference in spectral efficiency between the BV and MF model becomes smaller as the traffic grows and optical path demands become large.
As for the MF model, it outperforms the three other models in terms of required line cards, transponders, and BV-WSSs, while for fiber it takes second place. The equipment count reduction would contribute to energy-efficiency and simplify equipment management. The MLR model can save the number of line cards by aggregating 40/100 G transponders in a line card, while the MF model accommodates multiple low-rate optical paths in one transponder. Figure 5(d) shows that most IP layer connections in the MF model are 40 Gb/s or lower, if up to ten optical path demands are accommodated in an MFT. Figure 7 also shows that the MF model requires the fewest BV-WSSs, which is because the model uses BV-WSS ports most efficiently by combining multiple optical flows on one port. Contrarily, the MLR model requires more BV-WSSs because the model sometimes consumes more than one BV-WSS port on the add/drop block to accommodate an optical path demand that is even less than 400 Gb/s. Figure 9 shows the maximum BV-WSS port count in the cross-connect block among all (BV-)WXCs in the network. Decreases in the number of add/drop blocks leads to the reduction in BV-WSS port requirements in the cross-connect block.
4. Conclusions
In order to determine the cost-efficient multi-layer network architectures against the background of emerging elastic optical network concepts and technologies, we evaluated four IP over elastic optical network models that employ different types of transponders; Fixed, MLR, BV, and MF models. We modeled the (BV-)WXC, transponder, and router architecture and introduced the composition method to design multi-layer networks that are close to optimum. An evaluation based on static traffic demands showed that the MFT-based IP over elastic optical network is the lowest cost architecture because it accommodates multiple optical paths in an MFT, which reduces the required number of BV-WSSs, transponders, and router line cards. In addition, the model can mitigate the BV-WSS port requirements of (BV-)WXCs. In the model, most node pairs are connected directly by optical flows, while the interconnections between router and WXC are kept simple. This approach can mitigate the complexity of equipment management and enhance the feasibility of IP traffic offloading to the optical layer and the creation of energy-efficient networks. The cost assumption of MFT is unclear and further discussion is needed, but we believe that the MFT-based IP over elastic optical network would be one of promising architectures for future core networks. Evaluations of distance-adaptive modulation [9] and a survivability study that addresses transponder failure would be future study topics.
References and links
1. A. Klekamp, U. Gebhard, and F. Ilchmann, “Energy and cost efficiency of adaptive and mixed-line-rate IP over DWDM networks,” J. Lightwave Technol. 30(2), 215–221 (2012). [CrossRef]
2. M. Jinno, H. Takara, B. Kozicki, Y. Tsukishima, Y. Sone, and S. Matsuoka, “Spectrum-efficient and scalable elastic optical path network: architecture, benefits, and enabling technologies,” IEEE Commun. Mag. 47(11), 66–73 (2009). [CrossRef]
3. M. Jinno, H. Takara, Y. Sone, K. Yonenaga, and A. Hirano, “Multiflow optical transponder for efficient multilayer optical networking,” IEEE Commun. Mag. 50(5), 56–65 (2012). [CrossRef]
4. T. Tanaka, A. Hirano, and M. Jinno, “Performance evaluation of elastic optical networks with multi-flow optical transponders,” Proc. ECOC 2012, Tu.3.D.2 (2012). [CrossRef]
5. T. Tanaka, A. Hirano, and M. Jinno, “Impact of multi-flow transponder on equipment requirements in IP over elastic optical networks,” Proc. ECOC 2013, We.1.E.3 (2013).
6. S. Gringeri, B. Basch, V. Shukla, R. Egorov, and T. J. Xia, “Flexible architectures for optical transport nodes and networks,” IEEE Commun. Mag. 48(7), 40–50 (2010). [CrossRef]
7. G. J. Eilenberger, S. Bunse, L. Dembeck, U. Gebhard, F. Ilchmann, W. Lautenschlaeger, and J. Milbrandt, “Energy-efficient transport for the future internet,” Bell Labs Tech. J. 15(2), 147–167 (2010). [CrossRef]
8. A. Betker, C. Gerlach, R. Hülsermann, M. Jäger, M. Barry, S. Bodamer, J. Späth, C. Gauger, and M. Köhn, “Reference transport network scenarios,” Tech. Report, BMBF MultiTeraNet (2003).
9. M. Jinno, B. Kozicki, H. Takara, A. Watanabe, Y. Sone, T. Tanaka, and A. Hirano, “Distance-adaptive spectrum resource allocation in spectrum-sliced elastic optical path network,” IEEE Commun. Mag. 48(8), 138–145 (2010). [CrossRef]
10. F. Rambach, B. Konrad, L. Dembeck, U. Gebhard, M. Gunkel, M. Quagliotti, L. Serra, and V. López, “A multilayer cost model for metro/core networks,” J. Opt. Commun. Netw. 5(3), 210–225 (2013). [CrossRef]
11. O. Gerstel, M. Jinno, A. Lord, and S. J. B. Yoo, “Elastic optical networking: A new dawn for the optical layer?” IEEE Commun. Mag. 50(2), s12–s20 (2012). [CrossRef]
