Reliable Video Broadcasts via Protected Steiner Trees

Transcription

1 TOPICS IN OPTICAL COMMUNICATIONS Reliable Video Broadcasts via Protected Steiner Trees J. David Allen and Peter Kubat, Verizon Network and Technology ABSTRACT The introduction of the latest generation of s in the communication long-haul transport networks allows network planners to consider some new cost-effective design alternatives. Specifically, for video broadcast services wavelength drop-and-continue technology enables simple wavelength connections at each node via a tree-like topology, and the intelligent control plane permits the use of various shared protection schemes (with failure restoration switching times comparable to SONET BLSR). In this article we formulate models for reliable TV/video broadcast. We consider the network topologies based on minimum spanning trees. The objective is to minimize the total network cost while ensuring that the broadcast, originating in one (or two) source node(s), is delivered to a set of destination nodes and the network will tolerate at least one single link failure. The resulting protected tree networks are illustrated, and the cost of protection strategies is analyzed. INTRODUCTION National TV/video service providers that use terrestrial facilities (coaxial cable or fiber) for the last-mile distribution operate sophisticated, complex, and very reliable networks which deliver TV broadcast and video on demand (VoD) to the subscriber s home. These networks can be differentiated based on the lastmile transport technology. For example, coaxial cable is used by Comcast and Time- Warner, fiber to the home is used by Verizon, and a hybrid scheme a fiber to the curb, then a coaxial cable to the customer s home is used by AT&T. However, the long-haul video transport network architecture is almost identical for all [1]. Specifically, the live feeds and broadcast video originate in one, but most likely two identical super-head-ends (SHEs); the second SHE provides redundancy. From the SHE, the TV/video broadcast is carried over a long-haul (private) network to the regional video hub offices (s). In this application we assume that the video payload is carried in 10 Gb/s channels using one-toone protected wavelengths. All connections are via permanent circuits, and the locations of existing s remain static. At the, broadcast channels are repackaged, and local channels are added and further distributed (via one or more Gb/s channels imbedded in a 10 Gb/s wavelength) to the local video switching offices (VSOs). From the VSO, content is distributed to customers over the local access network [2]. VoD may be also added to the cable line-up or carried in a separate (data) channel (e.g., Verizon s FiOS TM ). The logical diagram of such architecture is illustrated in Fig. 1. We focus on the nationwide long-haul transport from the SHE to the s. Reconfigurable optical add-drop multiplexers (s) with drop-and-continue and split-and-continue have capabilities recently deployed in the longhaul networks [3] and permit dropping a single wavelength in multiple locations, allowing the broadcast distribution network to form a simple tree network topology. A single wavelength may, in principle, serve all s, but in this case the s would be only single-connected and thus vulnerable to a single fiber segment outage. To rectify this, we may require that all s be protected by a second wavelength where both wavelengths do not share a single common fiber segment (Fig. 2). We explore the use of Steiner minimum cost tree theory [4] in our tree-like architectures for the broadcast, then formulate models and finally propose a cost-minimal network solution. SHE- BROADCAST VIDEO DISTRIBUTION ARCHITECTURE In this application we assume that at each SHE, the live feeds and video broadcast streams from various satellites are coded/decoded, processed, compressed, and combined into a broadcast package containing many standard definition and high definition digital channels. This package is then assembled into a 10 Gb/s circuit and sent in a single wavelength over the long-haul optical network to the regional s. The current architecture (based on 1+1 synchronous optical network [SONET] rings) assigns two diversely routed wavelengths for the transport of broad /10/$ IEEE

2 cast signals to multiple s, essentially forming a ring riding in the dense wavelengthdivision multiplexed (DWDM) optical network systems. Each SHE has two rings extending to the s one ring for eastern s and another for western. This way each receives multiple copies of the video signal via SONET-protected transport and selects the best signal for distribution to the video subscribers. This has proven to be an extremely reliable but somewhat expensive architecture. The latest generation, managed via an intelligent control plane, offers planners alternative network architecture options. First, wavelength drop-and-continue technology enables simple wavelength connections at each node via a tree-like topology. Second, the intelligent control plane permits the use of various shared protection schemes for the wavelength (with failure restoration switching times comparable to SONET bidirectional line switched ring [BLSR]); thus, it may replace current expensive one-for-one wavelength redundancy. In the following sections we explore some cost-effective mesh network topologies suitable for reliable transport of unidirectional broadcast of TV/video. SHE SHE VSO VSO VSO VSO VSO Figure 1. Logical diagram of TV/video distribution networks. Long-haul network VSO Regionalmetro networks Local access networks SHE Figure 2. TV/video transport from SHE to : two non-overlapping trees. CONCEPT REVIEW AND RELATED WORK For packet networks, multicasting (inherently leading to a tree topology) has been extensively researched, resulting in a large body of literature. Optical multicasting for many demands in wavelength routed networks was suggested in [5]. The authors introduce the notion of lighttrees and apply it to broadcast/multicast. The network is then optimized for average packet hop distance and the total number of transceivers. In the subsequent paper [6] the light-tree concept is extended to include protection against a single link failure, and a heuristic based on a minimum cost Steiner tree is described. Wavelength engineering considerations for multicast optical routing is treated in [7]. A dual-tree concept for fault-tolerant multicast has also been proposed in [8]. The secondary tree, derived using simulation, functions as a backup to the primary tree with the goal of providing a short restoration time in the case of a primary tree link failure. In their recent paper Cha et al. [9] consider path protection routing with shared risk link group (SRLG) constraints in WDM mesh networks. In [9] the authors describe an IPTV application with a similar SHE- long-haul broadcast scenario, using an integer programming model to derive two minimum cost trees, each rooted in a different SHE, making sure that two diverse paths do not share a common SRLG. In [10] novel transport architectures to carry IPTV in metro networks have been proposed; these -based ring networks are relevant for regional (-VSO) TV/video broadcast. In our article the modeling and integer programming formulation, which is based on multicommodity flows, differs from all the above mentioned work and is uniquely suited for this type of TV/video distribution in which the circuits are permanent and the need for reliability is extremely high. The models described here are more specific and compact than in [9]; they handle not only a variety of dual-tree concepts but multitree architectures as well. Alternatives for even more reliable transport, such as two trees with fast reroute around a failed fiber segment, are also modeled. Long-haul network Minimize Cost: c ij x ij (0) Subject to: Flow conservation constraints for commodities; (1) Feasibility constraints: k f k ij N 0. x ij ; (2) Integrality constraints: x ij, f k ij = {0, 1}. (3) Figure 3. Steiner minimum tree formulation. 71

3 Step 1: Solve for a minimum Steiner tree (there will be m, m N 0 1 unidirectional links in the solution). /* This is the "working" tree. */ Step 2: For every link l ij in the Steiner minimum tree created in step 1, create a commodity of one unit demand from i to j. Step 3: Formulate and solve a standard Multi-Commodity Flow Problem (see [6] for details) with an added constraint set which prohibits use of link l ij by each commodity created in step 2. Step 4: Calculate the total cost of the solution, i.e., sum the cost of the Steiner tree and the cost of links kept in reserve as solved in step 3. Figure 4. Best one-failure protection algorithm. STEINER TREES, TWO-CONNECTED STEINER TREES, AND OTHER MODELS In DWDM optical systems equipped with nodes, a wavelength can be both dropped at a and still continue (after passing through optical amplifiers) to the next site without regeneration. Thus, in principle, one wavelength could reach and serve all the sites. In practice the number of sites may be limited due to the maximum distance a wavelength can travel, a restriction on the total number of drops, or regional restrictions, but these are not considered in this formulation. In our modeling effort we assume that we are given a connected network having N nodes (all equipped with s) connected with L links. Without loss of generality, the links are assumed to be unidirectional; l ij denotes the link from node i to node j. There is a node n s which is the source of the broadcast (the SHE is located there) and N 0 (< N) destination sites where the broadcast is dropped ( locations). The rest of the nodes (N N 0 ) can be involved in transmission, but the wavelength will either pass through or split (and continue) toward two or more neighbors. These nodes will be called Steiner nodes. Assume there is a cost c ij associated with each link l ij ( L ) if used in the solution. STEINER MINIMUM TREE FORMULATION A Steiner minimum tree problem can be described as follows: Select a subset of links L (L L) so that the source node n s is connected to the all destination nodes N 0 and the cost of (a) (b) Figure 5. Two trees destinations are dual-path protected: a) tree 1; b) tree 2; c) both trees. (c) 72

4 (a) (b) (c) (d) Figure 6. Three trees (multipath protected): a) tree 1 of 3 disjoint trees; b) tree 2 of 3 disjoint trees; c) tree 3 of 3 disjoint trees; d) all three. all selected links is minimized. The Steiner minimum tree can be formulated as a multicommodity flow integer programming (IP) problem [11]; its mathematical formulation is outlined in Fig. 3. Here, the commodity is defined as the TV/video broadcast content generated by the source. One unit of a virtual commodity is created for every source-destination pair (note there are N 0 commodities as each destination node must receive its own copy). Binary variables fij k represent the flow of the commodity k through the network: fij k = 1 if the flow k traverses link l ij, and fij k = 0 otherwise. The binary decision variable x ij describes link selection: x ij = 1 if the link l ij is selected, and x ij = 0 otherwise. Flow conservation constraints (1) make sure that one unit of commodity is routed continuously from its source to destination. The feasibility constraints (2) force the flows to go only over the links that have been selected by the algorithm, and the integrality constraints (3) prohibit splitting the flow and prohibit fractional link selection in the solution. In a typical application the number of locations (N 0 ) may be relatively small usually between sites. But in the national long-haul network, we can expect a large number of transmission nodes and consequently a large number of available transport segments. Thus, the size of the problem we are trying to solve may be very large and the solution algorithm challenging, as the Steiner minimum tree problem is NP-complete [12]. However, there are a number of commercially available integer programming solution packages that are able to solve these problems to optimality. Since the broadcast must be delivered with very high availability, single connectivity might not be enough; thus, protecting traffic using two different (and disjoint) wavelengths originating from two distinct SHEs is highly desirable. MINIMUM TWO-CONNECTED STEINER TREE FORMULATION We extend the Steiner minimum tree problem to provide double connectivity. A simple way to reformulate for double connectivity is to introduce a second set of commodities with flow variables gij k originating in the same SHE and terminating in the s, add duplicate constraints (1) and (2) for gij k, and add another constrain set: 73

5 Working routes Restoration routes Working routes Restoration routes (a) (b) Working routes Restoration routes (c) Figure 7. Restorable trees: a) restorable tree 1; b) restorable tree 2; c) two restorable trees. fij k + gk ij 1 for any commodity k and link ij. (4) Constraint set (4) guarantees that every destination node receives duplicate signals from the source, but by two completely non-overlapping paths. Again, the objective os the minimization of the total cost of links used in the solution. If a second source exists in a different location, then let the other commodity originate at the second source; the formulation remains the same. This scenario is, of course, more reliable. It is easy to see that this formulation can be further extended for greater levels of connectivity (e.g., three-connectivity). BEST ONE-FAILURE PROTECTION FOR A MINIMUM STEINER TREE Another approach for increasing availability of the broadcast long-haul network involves wavelength reroute in the case of a link failure. If spare channels in DWDM systems are available, the s intelligent control plane system will reroute the affected wavelength around the failed link. In this problem we want to minimize the total cost of (unidirectional) network segments that need to be kept in reserve so that every single link failure in the original working broadcast tree can be protected. Specifically, an alternative path can be constructed from these reserve links to avoid the failure and deliver the broadcast signal to all s. The best one failure protection algorithmic heuristic solution is outlined in Fig. 4. RESULTS The overall objective for the long-haul network is to ensure that each receives a signal from at least one SHE at all times, all at the minimum network cost. Several different spanning tree configurations were compared to evaluate the cost and benefits of using multiple redundant circuits and restorable circuits to provide the desired level of service availability. ILOG s OPLStudio 6.1 and CPLEX 11.2 were used to solve these spanning tree models to obtain the minimum-cost spanning tree connectivity in each case. These results were then compared to the existing 1+1 ring architecture to determine which architecture and configuration is preferable. In our study a normalized length of network segments was used as a proxy for the real cost in the interconnect facilities. 74

6 1 tree 2 disjoint trees 2 restorable disjoint trees 3 disjoint trees Rings Working fiber length* (W + R) Restoration fiber length* (Protect) Total fiber length* ,796 Fiber savings relative to rings 88% 73% 68% 60% N/A CPLEX run time (HH:MM) 00:01 00:03 *Fiber lengths are normalized Table 1. Summary. 00:18 (W + R1 + R2) 00:06 N/A The network used for these analyses is based on a real long-haul nationwide network, shown in the following figures. Triangles indicate locations of the SHEs, and the black dots mark the location of s; the lighter circles represent the Steiner (interconnection) nodes. The video network consists of two SHEs, each connected to 16 s via link-diverse spanning trees. While link+node diversity is possible, node diversity was not considered due to the extremely high node availability resulting from equipment redundancy at each node. Thus, overall service availability is dominated by link failures. Drop and continue was employed at s located at each and branching Steiner node in both spanning trees. Steiner nodes that do not perform a branching function merely pass traffic through without the need for a. Unidirectional 10 Gb/s transport was assumed on each spanning tree link. The first example of a spanning tree configuration, depicted in Figs. 5a and 5b, connects each of the two SHEs to every via a spanning (Steiner) tree, each rooted at different SHEs but not sharing a single common link (Fig. 5c). Here the total of the trees link lengths is minimized with the constraint that no unidirectional link in the network is used by more than one tree. If a link failure on one tree occurs, one or more s may be disconnected from the failed tree s SHE. However, the affected s would merely use the video signal being received from the other SHE s tree to maintain service. Further extensions are possible. For example, if the availability of the two-tree configuration is not deemed satisfactory, another disjoint tree could be added from a SHE to all of the s to provide additional redundancy. The three disjoint trees selected by the optimization model are shown in Fig. 6. Here, trees 1 (red) and 2 (green) reach all s, but tree 3 (blue) cannot connect to three of the s since these s are only 2-connected, so a 3rd disjoint path into these nodes is unavailable. So for all s other than these three, the network can survive at least two simultaneous link failures. In the third example (Fig. 7) we illustrate a one-failure protection scenario where restoration capacity (shown by dashed lines in the figures) is added to the working capacity to reroute traffic around a failed link. One advantage of this architecture is that the restoration capacity typically traverses existing fibers used by working links in the original trees (shown by solid lines). Thus, only additional restoration capacity is required to be added to an already in-use link, which is less expensive than lighting up a new link as is required by the original trees. Here, one option is to use only one restorable tree (red or green), whichever is less expensive, or use both for superior reliability albeit at higher cost. Table 1 summarizes the cost and benefits of each configuration illustration. Link distance is used as a proxy for link cost, and the fiber distance is normalized. The 1 tree case shows the total link length if each tree were optimized independently, i.e. not disjoint. The 2 disjoint trees column shows that using diverse trees requires 696 (16 percent) more fiber-miles than the sum of the tree lengths in the 1 tree case. This is the cost to ensure that the network can survive any single link failure. Adding restoration capacity to these two trees costs an additional 1000 fiber mi, as shown in the 2 restorable disjoint trees case. The disjoint trees used in the 3 disjoint trees case require a total length of 7446 mi. Note that all spanning tree cases use substantially less fiber miles than the rings case. Moreover, since the network cost can be assumed to be linearly proportional to the distance, we can expect substantial cost saving for the dual connected tree architecture. This savings is shown in the fourth row of data. The final row in the table shows the run time required to execute the model on a Sun Netra 440 DC server (UltraSparc IIIi) with 16 Gbytes RAM. CONCLUSIONS In this article we demonstrate the cost saving potential of the dual-connected tree-like architecture to deliver video broadcast reliably in the long-haul network. This approach 75

7 Our examples show that dual connected trees may reduce required miles of fiber by as much as 73 percent compared to current transport systems while providing the same or better reliability. has been motivated by the introduction of new generations of s with drop-andcontinue capability and intelligent control plane management. The new technology permits use of various shared wavelength protection schemes with failure restoration switching times comparable to SONET BLSR, making it suitable to replace current expensive one-for-one wavelength redundancy. Our examples show that dual connected trees may reduce required miles of fiber by as much as 73 percent compared to current transport systems while providing the same or better reliability. REFERENCES [1] Cisco, Cisco Wireline Video/IPTV Solution Design and Implementation Guide, Release 1.1, [2] M. Abrams et al., FTTP Deployment in the United States and Japan Equipment Choices and Service Provider Imperatives, IEEE J. Lightwave Tech., vol. 23, no. 1, Jan. 2005, pp [3] E. B. Basch et al., Architectural Tradeoffs for Reconfigurable Dense Wavelength-Division Multiplexing Systems, IEEE J. Sel. Topics Quantum Electronics, vol. 12, no. 4, 2006, pp [4] F. K. Hwang, D. S. Richards, and P. Winter, The Steiner Tree Problem, Elsevier, [5] L. H. Sahasrabuddhe and B. Mukherjee, Light-Trees: Optical Multicasting for Improved Performance in Wavelength-Routed Networks, IEEE Commun. Mag., vol. 36, no. 2, Feb. 1999, pp [6] N.K. Singhal et al., Provisioning of Survivable Multicast Sessions Against Single Link Failures in Optical WDM Mesh Networks, IEEE J. Lightwave Tech., vol. 21, no. 11, Nov. 2003, pp [7] Y. Xin and G. N. Rouskas, Multicast Routing Under Optical Layer Constraints, IEEE INFOCOM 04, vol. 4, Mar. 7 14, 2004, pp [8] A. Fei et al., A Dual-Tree Scheme for Fault-Tolerant Multicast, IEEE ICC 01, vol. 3, June 11 14, 2001, pp [9] M. Cha et al., Path Protection Routing with SRLG Constraints to Support IPTV in WDM Mesh Networks, IEEE INFOCOM 06, Apr , 2006, pp [10] S. Han, S. Lisle, and G. Nehib, IPTV Transport Architecture Alternatives and Economic Considerations, IEEE Commun. Mag., vol. 46, no. 2, Feb. 2008, pp [11] M. Pioro and D. Medhi, Routing, Flow, and Capacity Design in Communication and Computer Networks, Elsevier, [12] E. N. Gilbert and H.O. Pollak, Steiner Minimal Trees, SIAM J. Applied Mathematics, vol. 16, no. 1, 1968, pp BIBLIOGRAPHIES J. DAVID ALLEN (dave.allen1@verizon.com) is a Distinguished Member of Technical Staff in the Performance, Reliability, and Modeling Group of Verizon Network and Technology. His current work in Richardson, Texas, involves transport and architecture planning for Verizon s FiOS network. Prior to Verizon he spent 10 years with MCI and 15 years in the defense industry with Rockwell-Collins. He graduated in 1977 with a B.S.E.E. from Southern Methodist University (SMU) in Dallas, Texas, then went on to obtain an M.S. in telecommunications in 1980, an M.S. in operations research in 1984, and a D.Eng. degree in 1990, all from SMU. His current work interest is the application of Operations Research to the design and optimization of telecommunications networks. He also teaches an operations research survey course in SMU s Cox School of Business. PETER KUBAT [M 89] (peter.kubat@verizon.com) is a Distinguished Member of Technical Staff at the Performance, Reliability, and Modeling Group, Verizon Network and Technology, Waltham, Massachusetts. Currently, he is involved in network planning for optical/wdm networks and converged networks. Prior to joining Verizon he was a faculty member at the W. E. Simon Graduate School of Business Administration, University of Rochester. He graduated in 1969 from the Charles University of Prague (mathematics and physics), and in 1976 he received a D.Sc. degree in management science from the Technion Israel Institute of Technology. His current research interests include applied optimization, design and planning of optical/wdm and SONET networks, traffic engineering, and reliability of telecommunication systems. 76