Photonic MPLS Network Architecture Based on Hikari-Router [Invited]

Transcription

1 Photonic MPLS Network Architecture Based on Hikari-Router [Invited] Naoaki Yamanaka, Member, IEEE "IT Network Innovation Laboratories, Midori-cho, Musashino-shi, Tokyo, Japan yamanaka.naoaki@lab.ntt.co.jp Abstract - This paper describes multi-layer traffic engineering and signaling technologies in the photonic-gmpls-router (Hikari-Router) network. Multi-layer t 6 c engineering, which yields the dynamic cooperation of IP and photonic layers, is described with the goal of providing IP services costeffectively. The establishment of photonic cut-through paths is triggered when the Layer 3 traf ic loads become excessive which better uses the optical layer (Ll) resources. Data on tramic volumes are exchanged among edge photonic router using an extended IBGP. To realize multilayer traffic engineering, we propose an OSPF extension that advertises both the number of total wavelengths and the number of unreserved wavelengths, and an RSW-TE extension that minimizes the number of wavelength conversions needed. In addition, this paper presents a heuristic-based multi-layer topology design scheme in which IP traffiic measurements are performed in generalized multi-protocol label switches (GMPLSs). Our design scheme yields the optical label switch path (OLSP) network topology, i.e. OLSP placement that minimizes network cost in the face of fluctuations in IP traffic demand. In other words, the OLSP network topology is dynamically reconfigured to match IP traffic demand. Networks are reconfigured by the proposed scheme so as to utilize the network resources cost-effectively. I. INTRODUCTION The explosion of Internet traffic has led to a greater need for high-speed backbone networks. The growth in Internetprotocol (IP) traffic exceeds that of IP packet processing capability. Therefore, the next-generation backbone networks should consist of IP routers with IP-packet switching capability and optical cross-connects (OXC) with wavelength-path switching capability to reduce the burden of heavy IP-packet switching loads. In addition, IP traffic fluctuates often over periods as short as hours. These needs are satisfied by the photonic GMPLS router developed by "IT [l] that offers both IP packet switching and wavelength-path switching capabilities [2] [3]. Wavelength paths, called optical label switch paths (OLSP) are set and released in a distributed manner based on the functions offered by the generalized multi-protocol label switch (GMPLS). Since the photonic MPLS router offers the functions of both switching and GMPLS, it enables us to create the optimum network configuration by considering IP and photonic network resources in a distributed manner. An OSPF extension and a very sophisticated signaling mechanism that uses an extended form of RSW-TE have been proposed and will be implemented. The OSPF extension is to advertise both the total number of wavelengths and the number of unreserved wavelengths for each link. This information is passed via OSPF packets to all ingress edge nodes so they are aware of the GMPLS link state. Based on the least-load rule, ingress edge nodes try to establish more efficient OLSPs by using the extended RSW- TE signaling protocol. The signaling packet selects the wavelength that requires the least number of wavelength converters because wavelength conversion is a very expensive operation. In addition, multi-layer traffic engineering, which ensures that IP and photonic layers dynamically cooperate, is required to provide IP services cost-effectively. This paper proposes a heuristics-based multi-layer topology design scheme that uses IP traffic measurements. The proposed scheme yields the optimum OLSP network topology, i.e. OLSP placement, so as to minimize network cost, in response to fluctuations in IP traffic demand. In other words, OLSP network topology is dynamically reconfigured to match IP traffic demand [4]. JI.PHOTONIC GMPLS MULTI-LAYER ROUTER ARCHITEC- TURE (HIKARI-ROUTER) The photonic GMPLS multi-layer router consists of an IP router, wavelength router, and photonic-gmpls-router manager, as shown in Fig. 1 El]. The router is based on a photonic universal platform with the addition of 3R (Retiming, Reshaping and Regeneration) functions, wavelength conversion, and layer-3 functions. These functions are used adaptively. In other words, if the signal is degraded by fiber loss as well as nonlinear effects such as PMD (polarization mode dispersion) or ASE (Amplified spontaneous emission), the 3R function is activated. In addition, wavelength conversion is also used when signaling is blocked by wavelength overbooking. Of course, layer 3 (L3) packet forwarding is adaptively used when L3 packet forwarding is performed. Note that the photonic-gmpls router can transfer three types of signals. Type 1 is transparent transfer when the functions of 3R and wavelength conversion are not needed. Type 2 is & relay connections that need wavelength conversion. Type 2 connections are also supported by the adaptive use of the 3R function. Finally, type 3 involves layer 3 switching functions for operations such as MPLS path ag /03/$ IEEE 152

2 Fig. 1 Structure of photonic MPLS router with multi-layer traffic engineering based on IP traffic monitoring gregation and L3 packet level forwarding. The photonic- GMPLS router can support all these transfer capabilities. In the photonic-mpls-router manager, the GMPLS controller distributes own IP and photonic link states, and collects the link states of other photonic MPLS routers. IP traffic is always monitored, and the captured data is passed to the GMPLS controller through a low-pass filter [5]. A multi-layer topology design algorithm processes the collected IP and photonic link states and the collected traffic data. III. GMPLS MULTI-LAYER NETWORK CONTROL AND A TO- POLOGY DESIGN SCHEME A. Framework The bandwidth granularity of the photonic layer is coarse and equal to wavelength bandwidth, h, i.e. 2.5Gbit/s or logbit/s. On the other hand, the granularity of the IP layer is flexible and well engineered. Figure 2 shows that the center of the multi-layer network structure is a multi-layer photonic backbone network. This network is based on the multi-layer GMPLS router. At the edge node, all exchange traffic volumes between edge routers are monitored [5]. The volume of layer 3 traffic from edge #i to edge #j is exchanged by using the extended RSVP-TE protocol. The edge nodes use this information to establish the complete L3 traffic matrix. There is a trade-off between the Layer 3 hop-by-hop network and the photonic cut-through arrangement in terms of link cost and node cost as shown in Figure 3. When the traffic demand between source and destination IP routers is much less than h, the cut-through wavelength path between the source-destination IP routers is not fully utilized and so is not cost-effective. In this case, several IP traffic streams should be merged at some IP transit routers I Traffic matrix I Fig. 2 Multi-layer network design 153

3 I r Calculation model (a) Link Cost > Node Cost e Cost > Link Cost Fig. 3 Trade-off between Node Cost and Link Cost to better utilize the wavelength bandwidth at the cost of IPpacket processing at the transit nodes. On the other hand, when traffic demand between source and destination IP routers approaches or exceeds. h, such stream multiplexing is dropped. Thus, the setting of cut-through paths between sourcedestination IP routers depends on IP traffic, and the OLSP topology of the photonic layer should change dynamically according to fluctuations in IP-traffic demand to optimize network resource utilization. For this purpose, our objective is to minimize network cost 2, which is formulated as follows b0 Traffic demand between S-D pairs, T[Mbit/s] Fig. 4 Numerical model and optimum OLSP topology determined by IP traffic demand 0.6 U a, N 0.4 m " Traffic demand between S-D pairs, T[Mbit/s] Fig. 5 Cost reduction effect of reconfigurable multi-layer network. where cp is the cost of port p in router i, "jpk. is the cost of wavelength path k at port p in fiber link ij. u/b is the Noddink ratio. db is set to more than one, because IP routers have IP-packet processing functions such as table look up and packet-based switching in addition to wavelength routing functions. To minimize 2, we adopt the extended version of the BXCQ (branch exchange with quality-of-service constraints) scheme presented in [6], named EBXCQ. The BXCQ scheme was originally intended for multi-layer ATM network design. In BXCQ, the additiodelimination of links is iterated to solve a topological optimization problem with quality-of-service constraints, such as delay and blocking probability. In EBXCQ, the number of ports in both IP routers and wavelength routers, as well as the number of wavelengths per fiber are taken as additional constraints. B. Numerical results We have confirmed the effectiveness of EBXBQ by using a LATA network model, see Fig. 4 [7]. Figure 4 also shows that the optimum OLSP topology changes with the IP traffic demand between source-destination IP routers. We assume that IP traffic demands between sourcedestination IP routers are evenly distributed and that each wavelength is converted at each photonic MPLS router. Wavelength bandwidth is set to 2.5Gbitls. We used the metric of the average node degree, D, to characterize and evaluate OLSP network topologies. D is the average number of other IP routers to which individual IP routers are connected by OLSPs. As traffic demand increases, the optimum OLSP topology becomes a mesh because each OLSP bandwidth becomes more effectively utilized. Changing the OLSP topology in response to traffic demand fluctuations dramatically reduces network cost. An estimation of the network cost reduction possible is shown in Fig. 5. Network cost is normalized by the fixed optimum topology T=300 [Mbitls]. If the OLSP network topology is changed based on traffic measurements; the cost reduction can exceed 50%. Iv. ADAPTIVE TRANSPARENT PATH ARCHITECTURE 154

4 A. Architecture Transparent paths have the advantage of low cost because no O/E or E/O conversion is needed. However, they do suffer limited transmission distances. In addition, wavelength assignment is quite difficult. Figure 6 defines (a) Opaque and (b) Transparent optical paths. An opaque path uses O/E and E/O converters and so can offer wavelength conversion. This yields very flexible network design as well as route/wavelength selection. On the other hand, transparent paths have very low cost. Given today s technology, only amplifiers can be used in the transparent path. In other words, all optical wavelength conversion is not currently possible. We summarize the two optical path architectures in Table l. They are combined in the proposed adaptive transparent path technique, which is shown in Fig. 7. This approach tries to extend the transparent path for as long as possible. Details of the protocol are described in 4.2. This adaptive transparent network is realized with a wavelength conversion trunk as shown in Fig. 8. Note that the number of input line ports, K is larger than number of h-conversion units, k. In other words, the cost of E/O, O/E converter is much smaller than is true in the fully opaque network. Electrical switch I: Transparent (as long as possible) q-6- - A3 - OpLque Figure 7 Adaptive transparent path Optical SD switch (ex. PLC switch) h5 1 Transparent path (a) Opaque path Figure 8 Adaptive transparent optical path capable node architecture (b) Transparent path Fig. 6 Opaque and transparent paths in photonic network TABLE 1 CHARACTERISTICS OF TWO TYPES OF OPTICAL PATH bdvantages Disadvantages paque pailure detectionloe, E/O, Cost I B. Routing and signaling for proposed photonic GMPLS Currently available technologies make optical wavelength conversion very costly and a lot of hardware is needed. In other words, the transparent bit-rate restriction free network is very attractive from the viewpoints of low cost and high flexibility. Accordingly, our OSPF extension and RSVP-TE extension provide effective route selection as well as the least number of wavelength conversions. istance limit avelength assignment 155

5 ,' Type I, NW I I I I I length I I I I I NUW,. Signaling -, extension] i \ RSVP-TE Route A-B-C :,,,,,?xt,er;ls,iqn I 1 1 I I 1 I I Type I lenqth LRSVP-TE - I r lo : Unused wavelength 7 0: Used wavelength Unused wavelengths for both link A and B c NW: Number of wavelengths NUW: Number of unreserved wavelengths WC: Wavelength Converter 1 I If no unused wavelengths match use wavelength conversion Bb Figure 9 Photonic GMPLS extensions for both routing and signaling based on OSPF and RSVP-TE First, route advertisement is achieved by the OSPF extension as shown in Fig. 9. Each photonic GMPLS router advertises its total number of used and unused wavelengths. According to this advertisement, edge nodes can discern GMPLS link state and can select the least expensive path. In addition, we proposed to extend the sub-tlv of IETF specification. The extensions include 3R resource information and statistical information such as utilization of each wavelength, and 3R and wavelength conversion resources. This information is used for source routing based on a combination of shortest path first and load information. For the signaling function, we offer the RSVP-TE extension. Link by link routing is used based on source routing. The first photonic GMPLS router sets the unused wavelength information in bit map format in the RSVP signals. Each transit photonic GMPLS router over-writes this information by placing "And" between arriving signaling unused wavelength bitmap and its own unused wavelengths. If there is Electrical MPLS network..j..:.,.. ". "...:..."..."... 1 Adaptive wa\r 1 I E 0% 20% 40% 60% 80% 100% & Through traffic using OLSP Figure 10 Network cost reduction breakthrough no unused wavelength, wavelength conversion is used. The router, which offers wavelength conversion, creates a new unused wavelength bitmap and sends it to the next router vi- [91. This signaling and routing technique minimizes the fiequency of wavelength conversion in the network and so can provide very cost effective photonic networks. The calculation results of the proposed method cost reduction is shown in Fig. 10. The results show that cost reductions of more than 70% can be expected. V. CONCLUSIONS This paper presented multi-layer traffic engineering, signaling and routing technologies in the photonic-gmplsrouter (Hikari-router) networks. The photonic-gmplsrouter network architecture is based on OSPF and RSVP- TE extensions. The proposed routing and signaling technique minimizes the use of wavelength conversion and so yields very cost effective photonic networks. This paper also described a heuristic-based multi-layer topology design scheme, called EBXCQ, for IP photonic networks. By monitoring 1P traffic loads, photonic MPLS routers are controlled to dynamically change the network configuration; this allows network resources to be utilized efficiently in a distributed manner. REFERENCES [ 11 Y. Yamabayashi et al.,"autonomously Controlled Multiprotocol Wavelength Switching Network for Internet Backbones," IEICE Trans. Commun., vol. E83-B, no. 10, pp ,

6 K. Shimano et al., MPLambdaS Demonstration Employing Photonic Routers (256x256 OLSPS) To Integrate Optical and IP Networks, National Fiber Optic Engineers Conf Tech. Proc. p. 5. F. Kano et al., Development of Photonic Router (256x256 OLSPS) Realize Next Generation Network with MPLambdaS Control, National Fiber Optic Engineers Conf.2001 Tech. Proc. p Naoaki Yamanaka et al., Traffic Engineering and signaling technologies in photonic-mpls-router Network, Roc. on HPSR2002, Kobe, Japan, Session 3A- 4 K. Shiomoto, A Simple Bandwidth Management Strategy Based on Measurements of Instantaneous Virtual Path Utilization in ATM Networks, IEEE Trans. Networking, ~01.6, no. 5, pp , E. Oki and N. Yamanaka, Impact of Multimedia Traffic Characteristic on ATM Network Configuration, J. Network Systems Management, vol. 6, no. 4, pp ,1998. W. Imajuku et al., Multi-layer routing using multiplelayer switch capable LSRs, LETF draft, draft-imajukuml-routing-oo.txt, February (working in progress) E. Oki et al., Requirements of optical link-state information for traffic engineering, IETF draft, draft-okiipo-optlink-req-oo.txt, February (work in progress) N. Matsuura et al., Requirements for using RSVP-TE in GMPLS, IETF draft, draft-matsuura-gmpls-rsvprequirements-oo.txt, February (work in progress) 157