SingleLink Failure Detection in AllOptical Networks Using Monitoring Cycles and Paths


 April Pauline Park
 6 years ago
 Views:
Transcription
1 SingleLink Failure Detection in AllOptical Networks Using Monitoring Cycles and Paths Satyajeet S. Ahuja, Srinivasan Ramasubramanian, and Marwan Krunz Department of ECE, University of Arizona, Tucson, AZ 8571 {srini, Abstract In this paper, we consider the problem of fault localization in alloptical networks. We introduce the concept of monitoring cycles (MCs) and monitoring paths (MPs) for unique identification of singlelink failures. MCs and MPs are required to pass through one or more monitoring locations. They are constructed such that any singlelink failure results in the failure of a unique combination of MCs and MPs that pass through the monitoring location(s). For a network with only one monitoring location, we prove that threeedge connectivity is a necessary and sufficient condition for constructing MCs that uniquely identify any singlelink failure in the network. For this case, we formulate the problem of constructing MCs as an integer linear program (ILP). We also develop heuristic approaches for constructing MCs in the presence of one or more monitoring locations. For an arbitrary network (not necessarily threeedge connected), we describe a fault localization technique that uses both MPs and MCs and that employs multiple monitoring locations. We also provide a lineartime algorithm to compute the minimum number of required monitoring locations. Through extensive simulations, we demonstrate the effectiveness of the proposed monitoring technique. Index Terms Alloptical networks, fault localization. I. INTRODUCTION Optical networks have gained tremendous importance due to their ability to support very high data rates using the dense wavelength division multiplexing technology. With such high data rates, a brief service disruption in the operation of the network can result in the loss of a large amount of data. Commonly observed service disruptions are caused by fiber cuts, equipment failure, excessive bit errors, and human error. It is desired that these faults be uniquely identified and corrected at the physical layer before they are even noticed at higher layers. Therefore, it is critical for optical networks to employ fast and effective methods for identifying and locating network failures. Some failures, such as optical crossconnect port blocking and intrusion, can affect a single or a specific subset of wavelengths within a link. Other failures, including fiber cuts and high bit error rates (BERs), may affect all the wavelengths that pass through a fiber duct. In this work, we focus on the detection of the latter type of failures, and present a fault detection technique that can uniquely localize any singlelink failure. For ease of explanation, we use the notion of failure, although the treatment applies as well to assessing This work was supported by NSF under grants ANI , ANI , and Any opinions, findings, conclusions, and recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the NSF. other metrics that significantly impact the link performance, such as optical power, optical signaltonoise ratio (SNR), and BER. In order to rapidly measure the performance of a link (or a collection of links), it is essential to analyze the signal in the optical domain via optical spectrum analyzers (monitors). Failure detection in alloptical networks (AONs) may be performed by monitoring links individually, requiring a dedicated monitoring lightpath per link. The link quality in this case is assessed by transmitting specific test patterns. To assess the quality at various transmission rates, the test patterns must be transmitted several times at various rates. Instead of dedicating a monitoring wavelength per link, one may use inband approaches, whereby data channels (wavelengths) carry the test patterns in the overhead fields of the frame (similar to L1 monitoring in SONET/SDH networks). This requires electronic dropping of the data channel at the two ends of the link. In such a scenario, the data connection is no longer alloptical, and moreover it is not transparent (i.e., the enduser s framing procedure must be known to the network provider). Various opticallevel mechanisms for failure detection were proposed in the literature. These include optical spectral analysis, optical power detection, pilot tones, and optical time domain reflectometry (OTDR). In [6], a failure detection scheme was proposed, in which monitors are assigned to each optical multiplexing and transmission section. Such a scheme requires an excessive number of inherently expensive monitors, and is not scalable and cost effective for current AONs. In [13], a failure detection mechanism that employs independent wavelengths as supervisory channels was proposed. A cycle cover was used to monitor the links in the network. For each cycle, a node is designated to monitor that cycle s performance. Since the network is decomposed into cycles, faults can be localized at the cycle level, but this type of localization may not identify the failed link within that cycle. In [1], an adaptive technique for fault diagnosis using probes was presented in which probes are established sequentially, each time using information about already established probes. While the sequential probing helps achieve adaptiveness, it also increases the fault localization time. In [7], a nonadaptive fault diagnosis through a set of probes (lightpaths) was developed where all the probes are employed in advance. The techniques presented in [7] and [1] assume that a probe can originate from and terminate at any location. Such techniques can localize multiplelink failures, but they involve a high setup cost (a monitor is needed at each location) and incur high protocol overhead (for exchanging messages
2 between locations). By reducing the number of monitoring locations, the monitoring overhead and the detection delay can be significantly reduced because fewer messages are exchanged between monitoring locations. In this paper, we develop a mechanism for locating singlelink failures using monitoring cycles (MCs) only or a combination of MCs and monitoring paths (MPs). The MCs pass through one or more monitoring locations, and are selected such that the failure of any given link would result in the failure of a unique combination of MCs. We show that when one monitoring location is employed, finding a set of MCs that is sufficient to identify any singlelink failure requires the network to be threeedge connected. We provide upper and lower bounds on the number of MCs needed to uniquely identify any singlelink failure. The problem of constructing MCs in a threeedgeconnected network is formulated as an ILP, and a heuristic approach is presented to solve it. We then study the singlelink failure detection problem when multiple monitoring locations are employed. The effectiveness of employing multiple monitoring locations is compared with that of a single monitoring location. We then turn our attention to failure detection in an arbitrary topology (i.e., not necessarily threeedge connected). In this case, both MCs and MPs with multiple monitoring locations are needed to localize any singlelink failure. We provide necessary and sufficient conditions for the number of monitoring locations needed to identify all such failures. An O( L ) algorithm is presented for calculating the minimum number of required monitoring locations for a network with L links. Simulations are conducted to study the effectiveness of the monitoring technique. The rest of the paper is organized as follows. In Section II, we consider singlelink faultlocalization using MCs only and with one monitoring location. We show that threeedge connectivity of the network graph is a necessary and sufficient condition for the existence of a solution. We also provide theoretical bounds on the number of required MCs. In the same section, we present the ILP and heuristic solutions for the MCbased fault localization problem in a threeedgeconnected network. In Section III, we study the problem of identifying appropriate MCs when multiple monitoring locations are used, and propose a heuristic approach for solving it. In Section IV we study the problem of identifying singlelink failures in an arbitrary network and provide necessary and sufficient conditions on the number of monitoring locations. Section V presents the simulation results and finally the paper is concluded in Section VI. II. FAULT LOCALIZATION USING CYCLES AND ONE MONITORING LOCATION We first consider the scenario when the network employs one monitoring location. Given the monitoring location, the goal is to construct a set of MCs (or simply cycles) that pass through the monitoring location. The choice of the monitoring location depends on various factors such as its geographic location, security issues, and network topology. A wavelength is reserved for each cycle, and the failure of a cycle may be detected at the monitoring location using a dedicated monitor Links (1, ) (, 3) (3, 4) c 1 c c 3 c c c Links 1 3 (4, 1) (4, ) (1, 3) Fig. 1. Example network topology where node 1 is the monitoring location, monitoring cycles and its associated syndromes. per cycle or by time sharing the available set of monitors at the monitoring location. In addition to detecting link failures, MCs may also be used to assess various performance metrics, such as optical power, optical signaltonoise ratio (SNR), and BER. For example, if high BERs is observed on a link, the cycles traversing that link will also exhibit a high BERs. Since the combination of cycles is unique to a given link, the observed phenomena can be uniquely attributed to that link s performance. Problem MC1 (Monitoring Cycle Problem With One Monitoring Location): Given a network graph G(N, L), where N is the set of nodes and L is the set of undirected edges, and given a monitoring location m N. Find a set of MCs C such that every link in the network is present in at least one cycle and every singlelink failure results in the failure of a unique subset of cycles in C. All MCs must pass through the given monitoring location m. Such a unique subset is called the syndrome of the link. The set C is called as the faultdetection (FD) set. We refer to a set of MCs that can uniquely identify all singlelink failures as a feasible solution. In a feasible solution, every link must be present in at least one cycle, and hence a link failure results in the failure of at least one cycle. In addition, a link failure results in a unique syndrome. Consider the example in Figure 1, where node 1 is the monitoring location. A feasible solution is given by the three cycles c 1 = (1 4 1), c = ( ), and c 3 = ( ). The associated syndromes are also shown in Figure 1. For example, the syndrome for link (1, ) is the set of cycles {c 1, c 3 }. As every link is required to be on at least one cycle, the network must be at least twoedge connected 1 [],[3]. However, as we show next, threeedge connectivity is a necessary and sufficient condition for computing a feasible solution to the MC1 problem. Theorem 1: Threeedge connectivity is a necessary and sufficient condition for the existence of a solution to the MC1 problem. Proof: The necessity part is proved by contradiction. Assume to the contrary that there is a feasible solution to MC1 in a twoedge but not threeedgeconnected network. Since the network is not threeedge connected, there exists two links, say l and l, whose removal disconnects the network. Therefore, any cycle that traverses through one of these two links must also traverse the other. Thus, the failure of any of these two links cannot be uniquely identified through MCs, leading to a contradiction. Therefore, a feasible solution may be obtained only if any two link failures do not disconnect the 1 A kedgeconnected network is one in which the removal of any k 1 edges does not disconnect the network.
3 network, i.e., threeedge connectivity is a necessary condition for the existence of a solution. We now prove the sufficiency part by providing a method for constructing a feasible solution in a threeedgeconnected network. For any two links l = (x l, y l ) and l = (x l, y l ) in the network, we define three types of cycles: T 1 : Set of cycles containing l but not l. T : Set of cycles containing l but not l. T 3 : Set of cycles containing both l and l. To distinguish a failure between links l and l, it is obvious that a feasible solution must have cycles that belong to at least two of the above three types. For example, if feasible solution contains cycle c 1 T 1 and cycle c 3 T 3, then the failure of l will result in the failure of c 1 and c 3, whereas the failure of l will result in failure of c 3 only. Similarly, for other cycle combinations, i.e., (c 1 T 1, c T ) and (c T, c 3 T 3 ), we can easily show that the failures of l and l can be distinguished. In a threeedgeconnected network, one could easily construct cycles of type T 1 and T. A cycle of type T 1 may be constructed using the following two steps: (1) remove link l from the network (the resultant network is at least twoedge connected); and () in the resultant network, compute a cycle traversing the monitoring node m and link l, as shown in Figure. 1) Add a virtual node v and two virtual links (v, x l ) and (v, y l ). Remove link l. Note that the resultant graph is still twoedge connected. ) Obtain two linkdisjoint paths P 1 and P from v to m. 3) Form a cycle c by joining P 1 and P after removing the virtual links and adding the link l. Fig.. Procedure to compute a cycle traversing a given node m and link l in a twoedgeconnected network. A cycle of type T may be constructed using the same method above by interchanging l and l. Finally, to obtain a feasible solution, for each link pair (l, l ), we find cycles of types T 1 and T and take the union of all these cycles to form the FD set. In Theorem 1, we do not show an algorithm for constructing a cycle of type T 3. The FD set is constructed using cycles of type T 1 and T only. A. Bounds on the Required Number of Cycles A network with L links requires at least log ( L + 1) faultdetection cycles for any arbitrary choice of the monitoring node. This lower bound is achieved when all the edges are expressed as a binary representation of log ( L + 1) faultdetection cycles. We assume that at any time instance the network may have no failures or at most one link failure. In order to encode L +1 combinations in binary representations, we require log ( L + 1) bits, hence the number of cycles. For the example network of 6 links and the set of MCs presented in Figure 1, achieves the lower bound ( log (6+1) = 3). The sufficiency proof stated above provides a method for obtaining a feasible solution by constructing cycles of type T 1 and T for all possible link pairs. This results in an upper bound of L ( L 1) on the required number of cycles. A tighter bound can be obtained by modifying the construction of the feasible solution, as shown in Figure 3. Step ensures Step 1: Initialize: C φ. Step : l, l L (l l ) do 1) If C does not contain a cycle of types T 1 or T, then construct a cycle c of either type. ) C C {c}. Note: The above step adds at most L ( L 1)/ cycles. Step 3: l L do 1) If C does not contain any cycle involving link l, then construct a cycle c traversing m and link l. ) C C {c }. Fig. 3. ( ) L Construction of a feasible solution with + 1 cycles. that for every two links l and l, at least one cycle of type T 1 or T is present in C. Without loss of generality, assume that there exists a cycle c C such that c T 1. The maximum number of cycles added in this step is ( L ). There are two cases to consider: (1) There exists another cycle c that contains l, or () none of the cycles in C contains link l. In the former case, c T or c T 3. Therefore, C has cycles belonging to at least two of the three types. In the latter case, we add a cycle c that traverses the monitoring location m and link l (Step 3). Note that c must belong to T or T 3. Clearly, there can be at most one link in the network that belongs to this case. Hence, the upper bound on the number of cycles is ( L ) + 1. In Theorem 1, we do not present an algorithm for constructing cycles of type T 3. However, we argue that the exclusion of such cycles from the FD set does not change the upper bound on the size of the FD set. Now suppose that an algorithm exists for obtaining cycles of type T 3. We now argue that even when such cycles are used to construct a sufficient FD set, in the worstcase, the size of the FD set will still be ( ) L + 1, i.e., the T3 cycles have no impact on the upper bound. We start with an empty FD set. For each link pair (l, l ), suppose that we obtain a cycle of type T 3 and add it to the FD set. However, such a cycle cannot alone distinguish between the failures of l and l. Thus, to uniquely identify the failures of the two links, we still need another cycle of type T 1 or T. This is true for each link pair (l, l ). Hence, one cycle for each link pair is still required for failure identification. While the availability of T 3 cycles does not impact the upper bound ( ) L + 1, it provides more flexibility in selecting the FD set, which can lead to reducing the average cycle length in the solution set. In [7], a stricter bound on the number of monitoring probes was derived under the assumption that the probes can originate and terminate at any node in the network, i.e., If two links fall in this case, then Step would have ensured that at least one of the links is present in the solution.
4 the number of monitoring locations is not restricted. However, in our case, monitoring cycles originate from and terminate at the same monitoring location. B. Complexity For a threeedge connected network, a solution to MC1 problem can be obtained by using the procedure described in Section IIA which in the worst case involves O( L ) cycles. In [], it has been shown that the adaptive or sequential diagnosis problem is conpcomplete. We strongly believe that finding an optimal solution to the MC1 problem that minimizes the total network resource consumed belongs to the class of NPComplete problems. However, we do not give a formal proof for the same. C. Construction of an Optimal FD set From a hardware standpoint, building a monitoring system involves both operational and setup (onetime) cost. We assume that the operational cost per link is directly proportional to the number of wavelengths used for monitoring purposes over that link. Such dedicated wavelengths represent a loss of revenue for the network provider, as they would have otherwise been used to transport actual traffic. The setup cost is essentially the cost of monitors needed at the monitoring location. When monitors are dedicated to individual cycles, a wavelength is reserved over all the links in a cycle. The total operational cost is then proportional to the sum of the hop length of cycles employed in the monitoring system. The setup cost in this case is directly proportional to the number of monitoring cycles employed. If a network is heavily loaded, i.e., if the cost of reserving wavelengths over a period of time will dominate over the fixed cost associated with monitors, it is desired to minimize the sum of length of cycles employed for monitoring purpose. When monitors are timeshared, a single wavelength is reserved for monitoring purposes over each link, and only one monitor is employed at the monitoring location. In such a case, the operational cost is constant, i.e., one wavelength per link, and the setup cost is the cost of a monitor. To identify failures, cycles are sequentially scanned by the monitor. The larger the set of cycles, the higher the delay incurred in identifying a failure. Hence, when a single monitor is timeshared among a set of cycles, the objective should be to minimize the number of cycles in order to minimize the processing delay. In this work, we assume that a monitor is dedicated to each cycle and that the cost of reserving wavelengths over links dominate over the cost of monitors. We first formulate the problem of constructing a feasible solution that minimizes the sum of the cycle lengths as an ILP. We then develop heuristic solutions to it. To guarantee the existence of a solution, we assume that the network is threeedge connected. This restriction is relaxed later in Section IV. 1) ILP Formulation: For a network graph G(N, L) and a given monitoring location m N, let C m be the set of all cycles that pass through m. The objective of the ILP is to find an FD set such that the average number of cycles per link is minimized. The ILP formulation takes C m as input. Let α c be a binary variable that specifies if cycle c C m is part of the solution or not. Let L c be the hoplength of cycle c. For a link l L, let A lc be a binary indicator that denotes whether link l is present in cycle c or not; A lc = 1 if l L, and 0 otherwise. The ILP formulation is shown in Figure 4. The objective is to minimize the total number of hops consumed by the set of chosen cycles in G. This is equivalent to minimizing the sum of the lengths of the cycles. Constraint C1 ensures that for any two links l and l, there exists at least one cycle among the chosen cycles on which exactly one of the two links is present. Hence, the set of cycles that pass through a link is different from those that pass through all other links by at least one cycle. Since this is true for all link pairs, any link failure can be uniquely identified provided there is at least one cycle passing through the failed link. Constraint C ensures that every link is part of at least one cycle. Minimize c C m L c α c Subject to: C1: c C m α c (A lc (1 A l c) + (1 A lc )A l c) > 0, l, l L, l l C: c C m α c A lc > 0, l L Bounds: α c = {0, 1}, c C m Fig. 4. ILP formulation for finding the optimal FD set in a threeedgeconnected network. ) Efficient Heuristic for Constructing Monitoring Cycles: The above ILP solution guarantees finding the optimal FD set. However, its worstcase complexity grows exponentially with the number of integer variables in the formulation. Also, the algorithm requires as input all the cycles that pass through the monitoring node. This involves enumerating all cycles, so this step requires an exponentially large number of computations. The exponential complexity of the ILP approach makes it impractical for application in large networks. We now present a heuristic called MC1 for finding a good solution. The basic idea behind this algorithm is summarized in Figure 5. Step 1) Initialize C = φ and S = φ ) While L \ S φ a) Select a link l randomly from L \ S b) Compute cycle c passing through m and l using procedure in Figure. c) S = S {l : l c}, C = C {c} 3) While there exists a set of links L ( L ), where every link is present in the same set of cycles, a) Assign a large weight to all links in L and unit weight to all other links. b) Choose a random link l L c) Find a minimumweight cycle c through m and l using procedure in Figure. d) C = C {c} Fig. 5. Basic idea of the MC1 algorithm with node m as monitoring location. in Figure 5 ensures that each link is part of at least one cycle. Such a set of cycles is called the initial cover. To find the first cycle of the initial cover, a link l is chosen randomly
5 from the set L and a monitoring cycle passing through l is added to the FD set using the procedure presented in Figure. Further cycles are sequentially added to the solution set by randomly choosing a link l L that is not part of any cycle in the solution set, and then a cycle that passes through l is added to the solution set. Step 3 ensures that each link is associated with a unique combination of cycles. Let L denote the set of links that are present in the same set of cycles in C. A large weight is assigned to all the links in L. A link l is then chosen randomly from L and a cycle c passing through l is added to C using the procedure presented in Figure. It is guaranteed that cycle c does not contain at least one link from L \ {l } because the graph G is threeedge connected. Let us define L 1 = {l : l c L 1 } and L = L 1 \ L 1. The selection of cycle c guarantees that L 1 φ and L φ. It also guarantees that L 1 < L and L < L. Hence, by adding cycle c to the solution set, the MC1 algorithm splits L into two sets of smaller size. The above procedure is repeated until all links have unique syndrome. We maintain tag values for cycles and cumulative tag values for links to improve the running time of Step 3. The details of the tag values and other implementation details of the MC1 algorithm are presented in Appendix I. Correctness and Complexity: Assume to the contrary that the MC1 algorithm never terminates, i.e., the algorithm never comes out of the while loop in Step 3 in Figure 5. If it never terminates, then l and a set of n links {l (1), l (),..., l (n) } that are present in the same set of cycles. Since the MC1 algorithm never terminates, a shortest cycle for link l constructed using the procedure presented in Figure results in a cycle c that also passes through the set of links {l (1), l (),..., l (n) }. We show that this will never happen if the network is threeedge connected and that a shorter cycle c can be constructed. Recall that before adding a cycle, we set a large weight (say V ) to the n links {l (1), l (),..., l (n) }. The weight of cycle c (which is the sum of the weights of links in c) is then greater than nv. Now, remove from the network graph one of the n links say l (1). The graph is still twoedge connected. Finding a shortest cycle using the procedure in Figure will result in a cycle c of weight W (c ) < (n 1)V + L < W (c), because only n 1 links were assigned a large weight. The above is true provided that V L. Note that the addition of a cycle using the procedure in Figure results in distinguishing between the failures of at least two links. Hence, Step 3 requires a maximum of O( L ) iterations. A cycle traversing a node m and link l may be obtained in O( N log N + L ) time by running two instances of Dijkstra s algorithm [1]. The construction of an initial cover requires computing O( L ) cycles (in the worst case, one cycle per link) and Step 3 requires O( L ) cycles (one cycle for every link pair). Hence, the worstcase complexity of our algorithm is O( L ( N log N + L )). III. FAULT DETECTION USING MULTIPLE LOCATIONS We now consider a scenario where multiple locations are used to monitor singlelink failures in a threeedgeconnected network. This may be needed, for example, to provide more flexibility to the network operator, who may use multiple monitoring points for security and scalability reasons. Using more monitoring locations is also expected to result in a decrease in the average cycle length, and hence in faster localization of the failed link. Problem MCM (Monitoring Cycle Problem with Multiple Monitoring Locations): Given a graph G(N, L) and a set of monitoring locations M (with M = M), the goal is to find a set of MCs C such that every singlelink failure results in the failure of a unique combination of MCs. Each MC must pass through at least one of the monitoring locations. Depending on whether the monitoring locations share information about cycle failures or not, we define two modes of operation: 1. With information exchange: Monitoring locations exchange information about observed cycle failures to collaboratively localize a link failure. The information exchange process will add some delay to the failure localization time. However, as shown later, less network resources are needed for failure detection.. Without information exchange: The monitoring locations work independently and do not share any information. Fault localization is faster but may require additional network resources. In this case, every link failure results in the failure of a combination of cycles traversing a specific monitoring location. Our MCM algorithm for MC construction with multiple monitoring locations incorporates the two modes. For a given monitoring location, we define its cloud as the set of links that this location is responsible for monitoring. The main intuition behind the MCM algorithm is to choose cycles with relatively short lengths. The basic idea of the MCM algorithm can be described in the following two steps: 1. We associate each link with the cloud of the monitoring location that is closest to that link. By being closer to the monitoring location, it is more likely that the monitoring cycles passing through the link will be shorter. We refer to this step as cloud formation.. We find a set of cycles that pass through each monitoring location such that each link failure results in the failure of a unique combination of cycles: (1) in the network, if information exchange is employed; or () at a monitoring location, if information exchange is not employed. We refer to this step as cycle formation. When constructing cycles for the cloud of a given monitoring location, the MCM algorithm assigns very high weights to links that are not part of that cloud. This discourages the use of such links in forming the MCs and facilitates the construction of short cycles. Finally, the solution is further optimized by using cloud transfer. It is likely that when monitoring cycles for a cloud are added to the FD set, a set of links that are not part of this cloud will also result in unique syndromes. Such links can be removed from their respective clouds via a cloud transfer in order to conserve resources. Implementation details of the MCM algorithm are presented in Appendix II.
6 A. Monitoring with Paths and Cycles So far, we have used MCs to monitor singlelink failures. This constrains the solution space because every test signal that originates from a monitoring location must eventually return to it in order to close the loop. The use of paths along with cycles provides greater flexibility. A monitoring path (MP) originates from one monitoring location and terminates at another. The failure of a link along the path will be detected by the terminating monitoring location associated with that path. If the starting and ending monitoring locations are the same, the monitoring path reduces to a MC. For the example network and the feasible solution presented in Figure 1, if we replace cycle c 3 with the monitoring path p 1 = ( 4 3 1) and if we use node as a second monitoring location, we can still detect all singlelink failures. The associated syndromes generated with various link failures are given in Table I. Link c 1 c p 1 Link c 1 c p 1 Link c 1 c p 1 (1,) (4,1) (3,4) (,3) (4,) (1,3) TABLE I SYNDROMES ASSOCIATED WITH CYCLES c 1, c, AND PATH p 1. Before proceeding further, we formally define the monitoring problem with paths and cycles. Problem MPC (Monitoring with Paths and Cycles): Given a graph G(N, L) and a set of monitoring locations M, the goal is to find a set of monitoring paths P and a set of monitoring cycles C such that every singlelink failure results in the failure of a unique combination of paths and cycles in P C. We now describe a network transformation that converts the MPC problem into a modified version of the MC1 problem. First, we split each monitoring node u M into two virtual monitoring nodes u 1 and u, as shown in Figure 6. We introduce a central monitoring node J to which we connect the virtual monitoring nodes using auxiliary links. Finally, for each edge (u, v) L s.t. u M, we introduce auxiliary connector edges (u 1, v) and (u, v). We define the mapping T (.) such that T ((u 1, v)) = {(u, v)} and T ((u, v)) = {(u 1, v)}. Note that if both v and w are in M with (v, w) L, four auxiliary connector edges, (v 1, w 1 ), (v, w ), (v 1, w ), and (v, w 1 ), need to be added. In this case, T ((v i, w j )) def = {( v i, w j ), (v i, w j ), ( v i, w j )}, where i, j = {1, }, and v i and w i are, respectively, the toggles of v i and w i in the set {1,}. For example, T ((v 1, w 1 )) = {(v, w ), (v 1, w ), (v, w 1 )}. Node J is now the single monitoring location on the transformed graph. We now show that a solution to a slightly modified version of the MC1 problem with monitoring location at J in the transformed network will result in a solution to the MPC problem. Consider the cycle c = (J u 1 E F G v J) in Figure 6. If we eliminate the auxiliary edges, c becomes the path (u E F G v) in the original graph G. Similarly, cycle c = (J u 1 E F B u J) in the transformed network becomes cycle (u E F B u) in G. Notice that, an arbitrary cycle in the transformed network may not be a valid path or a cycle Monitoring Locations D u E H F B v G (a) Network before graph transformation. (b) Network after transformation. Fig. 6. Example of Network transformation presented in Section IIIA for monitoring a network using paths and cycles. in G. For example, the cycle c = (J u 1 E u J) in the transformed network is a selfloop (u E u) in G. Selfloops can be eliminated by removing some of the auxiliary connector links during cycle construction. Cycle Construction in the Transformed Network: For a twoedgeconnected graph, we can construct cycles containing link l = (x l, y l ) by adding a virtual node r, removing l, adding two virtual links (r, x l ) and (r, y l ), and finding two linkdisjoint paths from the monitoring location J to node r. Such paths can be obtained by assigning a unit capacity to each link in the network and then sequentially augmenting two units of flow from J to r along the two augmenting paths in the residual graph 3 [1] of the network. To avoid generating a selfloop in the original graph, after augmenting a unit flow along the first augmenting path P aug1, we remove the set of links T (l ) that corresponds to the auxiliary connector edge l P aug1. A selfloop can only be formed if T (l ) of an auxiliary connector edge l P aug1 is present in the residual graph. Notice that the removal of auxiliary connector edges from the residual graph will not disconnect the graph because the original graph is threeedge connected. The MC 1) Add a virtual node r and two virtual links (r, x l ) and (r, y l ). Remove link l. Note that the resultant graph is still twoedge connected. ) Augment a unit flow from J to r along the first augmenting path P aug1. 3) For every auxiliary connector edge l P aug1, remove T (l ) from the residual graph. 4) Obtain the second augmenting path P aug. 5) Construct two linkdisjoint paths P 1 and P by aggregating the flows of P aug1 and P aug. 6) Form a cycle c by joining P 1 and P after removing the virtual links and adding the link l. Fig. 7. Procedure to determine a cycle that traverses the monitoring location J and link l in the transformed network presented in Figure 6. 3 For a flow network G(N, L) in which each link (u, v) is associated with capacity c(u, v), and for a source s, sink t, and a flow f that is defined by a flow demand and a set of paths that support this demand, the residual network of G induced by f is G f (N, L f ), where L f = {(u, v) L : c f (u, v) > 0}. c f (u, v) is the amount of additional flow that we can push from u to v before exceeding c(u, v), and is given by c f (u, v) = c(u, v) f(u, v) [3]. For flow f, an augmenting path from s to t is any path on f with positive flow.
7 1 procedure presented in Section IIC. can now be applied to the transformed network with monitoring location at J. However, instead of calling the FindCycle procedure to determine a MC, the MC1 will now use the procedure in Figure 7. The L m input to MC1 in this case consists of all the links in the transformed graph except for the auxiliary links. The tag value (t l ) for an auxiliary connector edge l is set to the sum of tag values of cycles passing through l and T (l). IV. MONITORING IN ARBITRARY NETWORK TOPOLOGIES We have shown that threeedge connectivity is a necessary and sufficient condition to uniquely identify singlelink failures using one monitoring location. However, many optical networks are less than threeedge connected, and so they may require additional monitoring locations. We now show that with multiple monitoring locations and by employing MPs along with MCs, all the links of an arbitrarily connected network can be monitored. For any two links l and l, we classify paths into three categories: T 1 : Set of monitoring paths containing l but not l. T : Set of monitoring paths containing l but not l. T 3 : Set of monitoring paths containing both l and l. To distinguish the failures of links l and l, the FD set should contain paths of at least two of the above three types. We first derive the necessary conditions for monitoring a connected network using cycles and paths. Definitions: 1) The ksupergraph (or simply kgraph) of a network graph G(N, L) is a graph that represents the interconnectivity of the kedgeconnected components of G [8]. In other words, each node i of the kgraph is a kedgeconnected component of G. ) A component link l L is a link that connects two nodes in the same kedgeconnected component of G. 3) A connector link l L is a link that connects nodes belonging to two different kedgeconnected components of G. Such a link is also a link on the kgraph. For a given network G and a given integer k, the kgraph of G is unique [8]. The kgraph can be at most (k 1)edgeconnected [8], [5], [9]. In this paper, we employ  and 3 graphs to determine the minimum number of monitors required to uniquely localize all singlelink failures. Note that the  graph of a network is a tree. Lemma 1: To uniquely identify all singlelink failures in an arbitrary connected graph G(N, L), it is necessary that all the components of the graph that result from the removal of any one link l or two links l and l must each contain a monitoring location. Proof: Consider the removal of a link l from G. The removal of l will reduce the graph in to one or two connected components. If it results in one connected component, then this component must contain at least one monitoring location, because otherwise the network will be without any monitoring location. If the removal of l results in two components and only one component has a monitoring location, then no monitoring cycle or path can traverse l. Consider the removal of two links l and l from G. The removal of these links will reduce the graph into one, two, or three connected components. If it results in one connected component, then this component must contain at least one monitoring location, because otherwise the network will be without any monitoring location. If the removal of l and l results in two components (see Figure 8(a)) and only one of the components has a monitoring location, then any monitoring path or cycle that traverses l must traverse l. Thus, links l and l are not uniquely identifiable. Therefore, each component must contain at least one monitoring location. l l l C 1 C C 1 C C 3 l (a) Fig. 8. Decomposition of the network into connected components when two links l and l are removed. The removal results in: (a) two components and (b) three components. Suppose that the removal of l and l results in three components, as shown in Figure 8(b). The removal of link l results in two components and the removal of link l further disconnects one of the two components. As there cannot be any cycles traversing links l and l, the failure of these two links can be identified using monitoring paths only, thus requiring at least two monitoring locations. If component C 1 (C 3 ) does not contain a monitoring location, then an MP traversing l(l ) cannot be formed. Therefore, C 1 and C 3 must each have at least one monitoring location. Further, if C does not have a monitoring location, then every path from a monitoring location in C 1 to a monitoring location in C 3 will traverse links l and l. Hence, the failures of l and l cannot be distinguished. Therefore, C must also have a monitoring location. Corollary 1: To uniquely identify all singlelink failures in line and ring networks, all the nodes must be monitoring locations. Corollary : To uniquely identify all singlelink failures in a tree network, all the nodes of degree one or two must be monitoring locations. Corollary 3: In a twoedgeconnected network, all threeedgeconnected components with two or no connector edges must have a monitoring location. The proofs for the above three corollaries are obvious from Lemma 1. Minimum Number of Monitoring Locations and Their Placement: Using the lemma and corollaries described above, the minimum number of required monitoring locations in an arbitrary network can be obtained using the following procedure: In Steps 3 and 4 of Figure 9, the degree of a kedgeconnected component is that of the corresponding node on the kgraph, k =, 3. An example of a graph and its 3graph is shown in Figure. Notice that the graph of the example network is a tree. Each threeedgeconnected component of degree two or less in the example network has at least one monitoring location inside it. Similarly, each twoedgeconnected component has a monitoring location inside it. For example, component A is a twoedgeconnected component (b)
8 1) Obtain the graph of the original network graph G(N, L). ) Decompose every node in the graph (such a node is a edgeconnected component of the original graph) further into threeedgeconnected components, hence forming the 3graph of G. 3) Assign a monitoring location to every threeedgeconnected component of degree two or less. 4) Assign a monitoring location to every twoedgeconnected component of degree two or less, if such a component does not contain a monitoring location from step 3. Fig. 9. Procedure for finding the minimum number of monitoring locations and their placement. of the graph, and it is of degree two. Hence, it needs at least one monitoring location inside it. However, all threeedgeconnected components within A itself are of degree three and according to step 3 of Figure 9, they do not need a monitoring location. Hence, any arbitrary node is picked in step 4 as a monitoring location inside A. Similarly, component B is of degree two but a monitoring location is not assigned to it in step 4 because it already gets a monitoring location in one of its threeedgeconnected component based on step 3. Complexity: Unique decomposition of a graph into its  and Fig.. network. A B Monitoring node Edge node ThreeEdgeConnected Components TwoEdgeConnected Components Nonmonitoring node An example showing the graph and 3graph of an arbitrary 3graph can be done using O( L ) computations [8]. Identifying the degree of a component by identifying its outgoing edges can be done using O( L ) computations. Hence, the total complexity of the algorithm is O( L ). Correctness: Given the graph of a graph G and using Corollary, any feasible solution must have at least one monitoring location in each twoedgeconnected component of degree two or less. This is ensured by step 4 in Figure 9. Moreover, by using Corollary 3, a feasible solution that has the minimum possible number of monitoring locations must have a monitoring location inside each threeedgeconnected component of degree two, which is ensured by step 3. Finally, step 4 does not assign a monitoring location inside a twoedgeconnected component if there is at least one monitoring location already assigned inside it from step 3. Hence, the procedure in Figure 9 provides the minimum number of monitoring locations. C Let N 1 denote the number of twoedgeconnected components of degree two or less. Let N denote the number of threeedgeconnected components of degree two or less. Let N 3 denote the number of twoedgeconnected components of degree two or less that have threeedgeconnected components of degree two or less inside them. The minimum number of monitoring locations required in the network is then given by N 1 +N N 3. Note that the above procedure will place exactly one monitor if the given network is threeedge connected. We define an edge node in a twoedgeconnected component of a network as a node that is either a monitoring location or is part of a connector edge in the graph. Observe that every edge node is either a monitoring location or is connected to at least one monitor outside its twoedgeconnected component. By connected, we mean that there is a path that starts at the edge node and that terminates at a monitoring location outside the twoedgeconnected component. This path does not use any edges inside the twoedgeconnected component. Therefore, an edge node may be considered as a virtual monitor. For example, in Figure, there are two edge nodes inside component A. The edge node on the right side of component A is connected to the monitoring location in component C. Corollary 4: If a twoedgeconnected component of a connected network has exactly one edge node, then this component must be threeedge connected. In addition, the edge node must be a monitoring location. We now prove that the placement of monitors as described in Figure 9 is sufficient to uniquely identify all singlelink failures. As a first step, we show that there exists a monitoring path (or cycle) that traverses every link. Theorem : The placement procedure in Figure 9 will result in monitoring locations that are sufficient to construct MPs/MCs that can cover all the links. Proof: Consider a link l L. The removal of this link either disconnects the network or not. If it disconnects the network, then the two components connected by l must have a monitoring location (according to the necessary condition), and hence a path from a monitoring location in one component to that in the other must traverse link l. If the removal of link l does not disconnect the network, then l belongs to a twoedgeconnected component of the graph G. This component has one or more edge nodes. If it has one edge node, then by Corollary 4, the component is threeedge connected. Therefore, we can compute a MC traversing link l by using the procedure presented in Figure. If the twoedgeconnected component has more than one edge node, then two linkdisjoint paths can be computed from the virtual node to two distinct edge nodes. As every edge node is by itself a monitoring location or is connected to a monitoring location outside of the twoedgeconnected component, we can obtain an MP by extending the path between the edge nodes, if needed. Theorem 3: The monitoring locations placed according to the procedure in Figure 9 are sufficient to uniquely identify all singlelink failures. Proof: We prove the sufficiency condition by constructing a set of monitoring paths and cycles based on the monitoring
9 locations obtained using Figure 9. Theorem shows that for any link l there exists an MP that traverses this link. We now show that l is uniquely identifiable. After identifying the set of MPs that cover all the links, if two links l and l have a common syndrome, they are present in the same set of paths, i.e., a set of paths of type T 3. For these two links, we now show that there exists a path of type T 1 or T. We divide the links in the network into three sets. The first set (L 1 ) contains all links l L whose removal disconnects the graph. The second set (L ) contains all links l L whose removal does not disconnect the graph, but l L such that the removal of both l and l disconnects the graph. Finally, the third set (L 3 ) contains all other links. Each link in L 1 connects two nodes in the graph of G. Similarly, each link in L connects two nodes in the 3graph. Links in L 3 are contained within the nodes of the 3graph. Case 1: l L 1. We consider two subcases: (a) l L 1, and (b) l / L 1. Case 1(a): The removal of links l and l results in three components. Therefore, l connects two of these components. By the necessary condition, each of these two components must have a monitoring location. A path from one monitoring location in one component to a monitoring location in the other must traverse link l. Case 1(b): l is part of a twoedgeconnected component whose degree after the removal of link l is, say d. If d = 0 or 1, there exists a monitoring location in the component itself. As the component is twoedge connected 4, there exists two linkdisjoint paths from the monitoring location to the two nodes (x l and y l ) connected by link l. Thus, there exists a MC that traverses link l and not link l. If d > 1, then there exists a monitor along every edge node of the component. We can find a path between two edge nodes that traverses link l by replacing link l with a virtual node v and with two virtual links, and computing linkdisjoint paths from v to the two edge nodes. Therefore, there exists a path of type T 1 if l L 1. Case : l L. We consider three subcases: (a) l L 1, (b) l L, and (c) l L 3. Case (a): We can swap links l and l and use Case 1(b). Hence, the two links are uniquely identifiable. Case (b): The failure of the two links does not disconnect the graph. If the two links are present in different twoedgeconnected components, then the removal of l does not affect the twoedge connectivity of the component in which l is present. Therefore, similar to the argument in Case 1(b), we can find a monitoring path that traverses l and not l. If both l and l belong to L and their removal does not disconnect the graph, then the network can be decomposed into three components C 1, C, and C 3, as shown in Figure 11. There exists two links between C 1 and C of which one of them must be l. Similarly, there exists two links between C and C 3 of which one of them is l. According to the necessary condition, components C 1 and C 3 must have a monitoring location. Therefore, a path from a monitoring location in C 1 to a monitoring location in C 3 that traverses link l may be obtained by removing links l and l. 4 Note that the twoedge connectivity of the component is not disturbed by the removal of l because l L 1 C 1 l C l C 3 l l Fig. 11. Decomposition of a network into three components when l and l belong to L and their failure does not disconnect the network. Case (c): We swap the links and apply Case 3(b). Therefore, a monitoring path of type T exists. Case 3: l L 3. We consider two subcases: (a) l L 1, and (b) l / L 1. Case 3(a): We swap the links and apply Case 1(b). Thus, a monitoring path of type T exists. Case 3(b): The removal of l does not affect the twoedge connectivity of the component in which link l is present. Therefore, there exists a path between two distinct edge nodes in that component that traverses link l. If the component has only one edge node, then it is threeedge connected. Hence, there exists a MC that traverses link l after the removal of l. We now present an example of monitoring a network using paths and cycles. The network in Figure 1 has three threeedgeconnected components: C 1 = {1,, 3, 4}, C = {5, 6, 7, 8}, and C 3 = {9,, 11, 1}. Using the procedure in Figure 9, we present a solution to the monitoring problem with monitoring locations at nodes 7 and 11. It can easily be verified that N 1 =, N = 1, and N 3 = 1. The links in component C 1 are monitored by the two monitoring locations provisioned outside C 1, i.e., inside components C (node 7) and C 3 (node 11). The set of five paths and three cycles that result in unique syndromes for all links are p 1 = , p = , p 3 = , p 4 = , p 5 = , c 1 = , c = , c 3 = The set of singlelink failures and their corresponding syndromes are presented in Table II Fig. 1. Example of a network with monitoring locations at nodes 7 and V. SIMULATION RESULTS A. Results for the MC1 Problem We compare the performance of the MC1 algorithm with the optimal results obtained using the ILP solution. We first consider the four topologies in Figure 13. These small topologies are selected because they enable us to obtain the ILP solution using the CPLEX solver [4] in reasonable time. For the NJLATA and NSFNET topologies, we add a few additional links to make these topologies threeedge connected (a necessary condition for the existence of a solution using one monitoring location). For a given network topology, we vary the location of the monitoring node and report the average number of monitoringrelated wavelengths per link, given by c C L c/ L. This measure reflects the amount of resources per link needed for linkfailure detection. We also report the
Optical Layer Monitoring Schemes for Fast Link Failure Localization in AllOptical Networks
Optical Layer Monitoring Schemes for Fast Link Failure Localization in AllOptical Networks Bin Wu, PinHan Ho, Kwan L. Yeung, János Tapolcai and Hussein T. Mouftah Abstract Optical layer monitoring and
More informationApproximation Algorithms
Approximation Algorithms or: How I Learned to Stop Worrying and Deal with NPCompleteness Ong Jit Sheng, Jonathan (A0073924B) March, 2012 Overview Key Results (I) General techniques: Greedy algorithms
More informationMaximizing Restorable Throughput in MPLS Networks Reuven Cohen, Senior Member, IEEE, and Gabi Nakibly, Member, IEEE
568 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 18, NO. 2, APRIL 2010 Maximizing Restorable Throughput in MPLS Networks Reuven Cohen, Senior Member, IEEE, and Gabi Nakibly, Member, IEEE Abstract MPLS recovery
More informationMultilayer MPLS Network Design: the Impact of Statistical Multiplexing
Multilayer MPLS Network Design: the Impact of Statistical Multiplexing Pietro Belotti, Antonio Capone, Giuliana Carello, Federico Malucelli Tepper School of Business, Carnegie Mellon University, Pittsburgh
More informationThe Goldberg Rao Algorithm for the Maximum Flow Problem
The Goldberg Rao Algorithm for the Maximum Flow Problem COS 528 class notes October 18, 2006 Scribe: Dávid Papp Main idea: use of the blocking flow paradigm to achieve essentially O(min{m 2/3, n 1/2 }
More informationFairness in Routing and Load Balancing
Fairness in Routing and Load Balancing Jon Kleinberg Yuval Rabani Éva Tardos Abstract We consider the issue of network routing subject to explicit fairness conditions. The optimization of fairness criteria
More informationOn the TradeOff between Control Plane Load and Data Plane Efficiency in Software Defined Networks
1 Technion  Computer Science Department  Tehnical Report CS010  01 On the TradeOff between Control Plane Load and Data Plane Efficiency in Software Defined Networks Abstract Software Defined Networking
More informationA Network Flow Approach in Cloud Computing
1 A Network Flow Approach in Cloud Computing Soheil Feizi, Amy Zhang, Muriel Médard RLE at MIT Abstract In this paper, by using network flow principles, we propose algorithms to address various challenges
More informationThis article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.
IEEE/ACM TRANSACTIONS ON NETWORKING 1 A Greedy Link Scheduler for Wireless Networks With Gaussian MultipleAccess and Broadcast Channels Arun Sridharan, Student Member, IEEE, C Emre Koksal, Member, IEEE,
More information! Solve problem to optimality. ! Solve problem in polytime. ! Solve arbitrary instances of the problem. !approximation algorithm.
Approximation Algorithms Chapter Approximation Algorithms Q Suppose I need to solve an NPhard problem What should I do? A Theory says you're unlikely to find a polytime algorithm Must sacrifice one of
More informationDesign of LightTree Based Optical InterDatacenter Networks
Lin et al. VOL. 5, NO. 12/DECEMBER 2013/J. OPT. COMMUN. NETW. 1443 Design of LightTree Based Optical InterDatacenter Networks Rongping Lin, Moshe Zukerman, Gangxiang Shen, and WenDe Zhong Abstract Nowadays,
More informationPerformance Management and Fault Management. 1 Dept. of ECE, SRM University
Performance Management and Fault Management 1 Dept. of ECE, SRM University Performance Management Performance management requires monitoring of the performance parameters for all the connections supported
More informationIntegrated Backup Topology Control and Routing of Obscured Traffic in Hybrid RF/FSO Networks*
Integrated Backup Topology Control and Routing of Obscured Traffic in Hybrid RF/FSO Networks* Abhishek Kashyap, Anuj Rawat and Mark Shayman Department of Electrical and Computer Engineering, University
More informationOn Optimal Monitor Placement for Localizing Node Failures via Network Tomography
On Optimal Monitor Placement for Localizing Node Failures via Network Tomography Liang Ma, Ting He, Ananthram Swami, Don Towsley, and Kin K. Leung IBM T. J. Watson Research Center, Yorktown, NY, USA. Email:
More informationChapter 11. 11.1 Load Balancing. Approximation Algorithms. Load Balancing. Load Balancing on 2 Machines. Load Balancing: Greedy Scheduling
Approximation Algorithms Chapter Approximation Algorithms Q. Suppose I need to solve an NPhard problem. What should I do? A. Theory says you're unlikely to find a polytime algorithm. Must sacrifice one
More informationSecurityAware Beacon Based Network Monitoring
SecurityAware Beacon Based Network Monitoring Masahiro Sasaki, Liang Zhao, Hiroshi Nagamochi Graduate School of Informatics, Kyoto University, Kyoto, Japan Email: {sasaki, liang, nag}@amp.i.kyotou.ac.jp
More informationPerformance of networks containing both MaxNet and SumNet links
Performance of networks containing both MaxNet and SumNet links Lachlan L. H. Andrew and Bartek P. Wydrowski Abstract Both MaxNet and SumNet are distributed congestion control architectures suitable for
More informationAdaptive Linear Programming Decoding
Adaptive Linear Programming Decoding Mohammad H. Taghavi and Paul H. Siegel ECE Department, University of California, San Diego Email: (mtaghavi, psiegel)@ucsd.edu ISIT 2006, Seattle, USA, July 9 14, 2006
More informationOutline. NPcompleteness. When is a problem easy? When is a problem hard? Today. Euler Circuits
Outline NPcompleteness Examples of Easy vs. Hard problems Euler circuit vs. Hamiltonian circuit Shortest Path vs. Longest Path 2pairs sum vs. general Subset Sum Reducing one problem to another Clique
More informationDetermine: route for each connection and protect them if necessary to minimize total network cost (say wavelengthlinks).
Service Provisioning to Provide PerConnectionBased Availability Guarantee in WDM Mesh Networks Jing Zhang, Keyao Zhu, Hui Zang, and Biswanath Mukherjee Abstract We present availability analysis for WDMmeshnetwork
More informationIntegrating Benders decomposition within Constraint Programming
Integrating Benders decomposition within Constraint Programming Hadrien Cambazard, Narendra Jussien email: {hcambaza,jussien}@emn.fr École des Mines de Nantes, LINA CNRS FRE 2729 4 rue Alfred Kastler BP
More informationA Sublinear Bipartiteness Tester for Bounded Degree Graphs
A Sublinear Bipartiteness Tester for Bounded Degree Graphs Oded Goldreich Dana Ron February 5, 1998 Abstract We present a sublineartime algorithm for testing whether a bounded degree graph is bipartite
More informationEfficient Alarm Management in Optical Networks
Efficient Alarm Management in Optical Networks Sava Stanic, Suresh Subramaniam, Hongsik Choi, Gokhan Sahin, and HyeongAh Choi Department of Electrical & Computer Engineering Department of Computer Science
More informationA Practical Scheme for Wireless Network Operation
A Practical Scheme for Wireless Network Operation Radhika Gowaikar, Amir F. Dana, Babak Hassibi, Michelle Effros June 21, 2004 Abstract In many problems in wireline networks, it is known that achieving
More information6.852: Distributed Algorithms Fall, 2009. Class 2
.8: Distributed Algorithms Fall, 009 Class Today s plan Leader election in a synchronous ring: Lower bound for comparisonbased algorithms. Basic computation in general synchronous networks: Leader election
More informationProbe Station Placement for Robust Monitoring of Networks
Probe Station Placement for Robust Monitoring of Networks Maitreya Natu Dept. of Computer and Information Science University of Delaware Newark, DE, USA, 97 Email: natu@cis.udel.edu Adarshpal S. Sethi
More informationTHE last two decades have witnessed an exponential
IEEE JSAC  SAMPLING 2006 1 Practical Beacon Placement for Link Monitoring Using Network Tomography Ritesh Kumar and Jasleen Kaur Abstract Recent interest in using tomography for network monitoring has
More informationLoad Balancing and Switch Scheduling
EE384Y Project Final Report Load Balancing and Switch Scheduling Xiangheng Liu Department of Electrical Engineering Stanford University, Stanford CA 94305 Email: liuxh@systems.stanford.edu Abstract Load
More informationDiscuss the size of the instance for the minimum spanning tree problem.
3.1 Algorithm complexity The algorithms A, B are given. The former has complexity O(n 2 ), the latter O(2 n ), where n is the size of the instance. Let n A 0 be the size of the largest instance that can
More informationLOAD BALANCING IN WDM NETWORKS THROUGH DYNAMIC ROUTE CHANGES
LOAD BALANCING IN WDM NETWORKS THROUGH DYNAMIC ROUTE CHANGES S.Ramanathan 1, G.Karthik 1, Ms.G.Sumathi 2 1 Dept. of computer science Sri Venkateswara College of engineering, Sriperumbudur, 602 105. 2 Asst.professor,
More informationWIRELESS communication channels have the characteristic
512 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 54, NO. 3, MARCH 2009 EnergyEfficient Decentralized Cooperative Routing in Wireless Networks Ritesh Madan, Member, IEEE, Neelesh B. Mehta, Senior Member,
More informationTHE PROBLEM WORMS (1) WORMS (2) THE PROBLEM OF WORM PROPAGATION/PREVENTION THE MINIMUM VERTEX COVER PROBLEM
1 THE PROBLEM OF WORM PROPAGATION/PREVENTION I.E. THE MINIMUM VERTEX COVER PROBLEM Prof. Tiziana Calamoneri Network Algorithms A.y. 2014/15 2 THE PROBLEM WORMS (1)! A computer worm is a standalone malware
More information17.3.1 Follow the Perturbed Leader
CS787: Advanced Algorithms Topic: Online Learning Presenters: David He, Chris Hopman 17.3.1 Follow the Perturbed Leader 17.3.1.1 Prediction Problem Recall the prediction problem that we discussed in class.
More information! Solve problem to optimality. ! Solve problem in polytime. ! Solve arbitrary instances of the problem. #approximation algorithm.
Approximation Algorithms 11 Approximation Algorithms Q Suppose I need to solve an NPhard problem What should I do? A Theory says you're unlikely to find a polytime algorithm Must sacrifice one of three
More information8.1 Min Degree Spanning Tree
CS880: Approximations Algorithms Scribe: Siddharth Barman Lecturer: Shuchi Chawla Topic: Min Degree Spanning Tree Date: 02/15/07 In this lecture we give a local search based algorithm for the Min Degree
More informationDynamic CongestionBased Load Balanced Routing in Optical BurstSwitched Networks
Dynamic CongestionBased Load Balanced Routing in Optical BurstSwitched Networks Guru P.V. Thodime, Vinod M. Vokkarane, and Jason P. Jue The University of Texas at Dallas, Richardson, TX 750830688 vgt015000,
More informationAn Efficient PrimarySegmented Backup Scheme for Dependable RealTime Communication in Multihop Networks
IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 11, NO. 1, FEBRUARY 2003 81 An Efficient PrimarySegmented Backup Scheme for Dependable RealTime Communication in Multihop Networks Krishna Phani Gummadi, Madhavarapu
More informationMinimizing Probing Cost and Achieving Identifiability in Probe Based Network Link Monitoring
Minimizing Probing Cost and Achieving Identifiability in Probe Based Network Link Monitoring Qiang Zheng, Student Member, IEEE, and Guohong Cao, Fellow, IEEE Department of Computer Science and Engineering
More informationWhy? A central concept in Computer Science. Algorithms are ubiquitous.
Analysis of Algorithms: A Brief Introduction Why? A central concept in Computer Science. Algorithms are ubiquitous. Using the Internet (sending email, transferring files, use of search engines, online
More informationCompetitive Analysis of On line Randomized Call Control in Cellular Networks
Competitive Analysis of On line Randomized Call Control in Cellular Networks Ioannis Caragiannis Christos Kaklamanis Evi Papaioannou Abstract In this paper we address an important communication issue arising
More informationAN OVERVIEW OF QUALITY OF SERVICE COMPUTER NETWORK
Abstract AN OVERVIEW OF QUALITY OF SERVICE COMPUTER NETWORK Mrs. Amandeep Kaur, Assistant Professor, Department of Computer Application, Apeejay Institute of Management, Ramamandi, Jalandhar144001, Punjab,
More informationThe fatstack and universal routing in interconnection networks
The fatstack and universal routing in interconnection networks Kevin F. Chen, Edwin H.M. Sha Department of Computer Science, University of Texas at Dallas, Richardson, TX 75083, USA Abstract This paper
More informationOptimizing the Placement of Integration Points in Multihop Wireless Networks
Optimizing the Placement of Integration Points in Multihop Wireless Networks Lili Qiu, Ranveer Chandra, Kamal Jain, and Mohammad Mahdian ABSTRACT Efficient integration of a multihop wireless network
More informationCMSC 858T: Randomized Algorithms Spring 2003 Handout 8: The Local Lemma
CMSC 858T: Randomized Algorithms Spring 2003 Handout 8: The Local Lemma Please Note: The references at the end are given for extra reading if you are interested in exploring these ideas further. You are
More informationConnected Identifying Codes for Sensor Network Monitoring
Connected Identifying Codes for Sensor Network Monitoring Niloofar Fazlollahi, David Starobinski and Ari Trachtenberg Dept. of Electrical and Computer Engineering Boston University, Boston, MA 02215 Email:
More informationApplied Algorithm Design Lecture 5
Applied Algorithm Design Lecture 5 Pietro Michiardi Eurecom Pietro Michiardi (Eurecom) Applied Algorithm Design Lecture 5 1 / 86 Approximation Algorithms Pietro Michiardi (Eurecom) Applied Algorithm Design
More informationeach college c i C has a capacity q i  the maximum number of students it will admit
n colleges in a set C, m applicants in a set A, where m is much larger than n. each college c i C has a capacity q i  the maximum number of students it will admit each college c i has a strict order i
More informationImpact Of Interference On Multihop Wireless Network Performance
Impact Of Interference On Multihop Wireless Network Performance Kamal Jain Jitendra Padhye Venkata N. Padmanabhan Lili Qiu Microsoft Research One Microsoft Way, Redmond, WA 98052. {kamalj, padhye, padmanab,
More informationComputer Network. Interconnected collection of autonomous computers that are able to exchange information
Introduction Computer Network. Interconnected collection of autonomous computers that are able to exchange information No master/slave relationship between the computers in the network Data Communications.
More informationLoad Balancing. Load Balancing 1 / 24
Load Balancing Backtracking, branch & bound and alphabeta pruning: how to assign work to idle processes without much communication? Additionally for alphabeta pruning: implementing the youngbrotherswait
More informationJUSTINTIME SCHEDULING WITH PERIODIC TIME SLOTS. Received December May 12, 2003; revised February 5, 2004
Scientiae Mathematicae Japonicae Online, Vol. 10, (2004), 431 437 431 JUSTINTIME SCHEDULING WITH PERIODIC TIME SLOTS Ondřej Čepeka and Shao Chin Sung b Received December May 12, 2003; revised February
More informationNetwork (Tree) Topology Inference Based on Prüfer Sequence
Network (Tree) Topology Inference Based on Prüfer Sequence C. Vanniarajan and Kamala Krithivasan Department of Computer Science and Engineering Indian Institute of Technology Madras Chennai 600036 vanniarajanc@hcl.in,
More informationEmbedded Systems 20 BF  ES
Embedded Systems 201  Multiprocessor Scheduling REVIEW Given n equivalent processors, a finite set M of aperiodic/periodic tasks find a schedule such that each task always meets its deadline. Assumptions:
More informationPartial Path Protection for WDM Networks: EndtoEnd Recovery Using Local Failure Information
Partial Path Protection for WDM Networks: EndtoEnd Recovery Using Local Failure Information Hungjen Wang, Eytan Modiano, Muriel Médard Laboratory for Information and Decision Systems Massachusetts Institute
More informationBranchandPrice Approach to the Vehicle Routing Problem with Time Windows
TECHNISCHE UNIVERSITEIT EINDHOVEN BranchandPrice Approach to the Vehicle Routing Problem with Time Windows Lloyd A. Fasting May 2014 Supervisors: dr. M. Firat dr.ir. M.A.A. Boon J. van Twist MSc. Contents
More informationBounded Cost Algorithms for Multivalued Consensus Using Binary Consensus Instances
Bounded Cost Algorithms for Multivalued Consensus Using Binary Consensus Instances Jialin Zhang Tsinghua University zhanggl02@mails.tsinghua.edu.cn Wei Chen Microsoft Research Asia weic@microsoft.com Abstract
More informationA Scalable Monitoring Approach Based on Aggregation and Refinement
IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL 20, NO 4, MAY 2002 677 A Scalable Monitoring Approach Based on Aggregation and Refinement YowJian Lin, Member, IEEE and Mun Choon Chan, Member, IEEE
More informationEmbedded Systems 20 REVIEW. Multiprocessor Scheduling
Embedded Systems 0   Multiprocessor Scheduling REVIEW Given n equivalent processors, a finite set M of aperiodic/periodic tasks find a schedule such that each task always meets its deadline. Assumptions:
More informationLabeling outerplanar graphs with maximum degree three
Labeling outerplanar graphs with maximum degree three Xiangwen Li 1 and Sanming Zhou 2 1 Department of Mathematics Huazhong Normal University, Wuhan 430079, China 2 Department of Mathematics and Statistics
More informationCROSS LAYER BASED MULTIPATH ROUTING FOR LOAD BALANCING
CHAPTER 6 CROSS LAYER BASED MULTIPATH ROUTING FOR LOAD BALANCING 6.1 INTRODUCTION The technical challenges in WMNs are load balancing, optimal routing, fairness, network autoconfiguration and mobility
More informationPROTECTION ALGORITHMS FOR BANDWIDTH GUARANTEED CONNECTIONS IN MPLS NETWORKS WONG SHEK YOON
PROTECTION ALGORITHMS FOR BANDWIDTH GUARANTEED CONNECTIONS IN MPLS NETWORKS WONG SHEK YOON (B.Eng.(Hons), NUS) A THESIS SUBMITTED FOR THE DEGREE OF MASTER OF ENGINEERING DEPARTMENT OF ELECTRICAL & COMPUTER
More informationBeyond the Stars: Revisiting Virtual Cluster Embeddings
Beyond the Stars: Revisiting Virtual Cluster Embeddings Matthias Rost Technische Universität Berlin September 7th, 2015, TélécomParisTech Joint work with Carlo Fuerst, Stefan Schmid Published in ACM SIGCOMM
More informationIN THIS PAPER, we study the delay and capacity tradeoffs
IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 15, NO. 5, OCTOBER 2007 981 Delay and Capacity TradeOffs in Mobile Ad Hoc Networks: A Global Perspective Gaurav Sharma, Ravi Mazumdar, Fellow, IEEE, and Ness
More informationMax Flow, Min Cut, and Matchings (Solution)
Max Flow, Min Cut, and Matchings (Solution) 1. The figure below shows a flow network on which an st flow is shown. The capacity of each edge appears as a label next to the edge, and the numbers in boxes
More informationAlgorithm Design and Analysis
Algorithm Design and Analysis LECTURE 27 Approximation Algorithms Load Balancing Weighted Vertex Cover Reminder: Fill out SRTEs online Don t forget to click submit Sofya Raskhodnikova 12/6/2011 S. Raskhodnikova;
More informationNetworked Computing in Wireless Sensor Networks for Structural Health Monitoring
Networked Computing in Wireless Sensor Networks for Structural Health Monitoring Apoorva Jindal, Member, IEEE, Mingyan Liu, Senior Member, IEEE, Abstract This paper studies the problem of distributed computation
More informationAnalysis of Algorithms, I
Analysis of Algorithms, I CSOR W4231.002 Eleni Drinea Computer Science Department Columbia University Thursday, February 26, 2015 Outline 1 Recap 2 Representing graphs 3 Breadthfirst search (BFS) 4 Applications
More informationCapacity planning and.
Some economical principles Hints on capacity planning (and other approaches) Andrea Bianco Telecommunication Network Group firstname.lastname@polito.it http://www.telematica.polito.it/ Assume users have
More informationA New Fault Tolerant Routing Algorithm For GMPLS/MPLS Networks
A New Fault Tolerant Routing Algorithm For GMPLS/MPLS Networks Mohammad HossienYaghmae Computer Department, Faculty of Engineering, Ferdowsi University of Mashhad, Mashhad, Iran hyaghmae@ferdowsi.um.ac.ir
More informationOptimal Index Codes for a Class of Multicast Networks with Receiver Side Information
Optimal Index Codes for a Class of Multicast Networks with Receiver Side Information Lawrence Ong School of Electrical Engineering and Computer Science, The University of Newcastle, Australia Email: lawrence.ong@cantab.net
More informationSystem Interconnect Architectures. Goals and Analysis. Network Properties and Routing. Terminology  2. Terminology  1
System Interconnect Architectures CSCI 8150 Advanced Computer Architecture Hwang, Chapter 2 Program and Network Properties 2.4 System Interconnect Architectures Direct networks for static connections Indirect
More informationJOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 9, MAY 1, 2009 1
JOURNAL OF LIGHTWAVE TECHNOLOGY, VOL. 27, NO. 9, MAY 1, 2009 1 Monitoring Cycle Design for Fast Link Failure Localization in AllOptical Networks Bin Wu, Member,, Kwan L. Yeung, Senior Member,, and PinHan
More informationInformation Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay
Information Theory and Coding Prof. S. N. Merchant Department of Electrical Engineering Indian Institute of Technology, Bombay Lecture  17 ShannonFanoElias Coding and Introduction to Arithmetic Coding
More informationMonitor Placement for Maximal Identifiability in Network Tomography
Monitor Placement for Maximal Identifiability in Network Tomography Liang Ma, Ting He, Kin K. Leung, Ananthram Swami, and Don Towsley Imperial College, London, UK. Email: {l.ma10, kin.leung}@imperial.ac.uk
More informationarxiv:cs/0605141v1 [cs.dc] 30 May 2006
General Compact Labeling Schemes for Dynamic Trees Amos Korman arxiv:cs/0605141v1 [cs.dc] 30 May 2006 February 1, 2008 Abstract Let F be a function on pairs of vertices. An F labeling scheme is composed
More informationNetwork File Storage with Graceful Performance Degradation
Network File Storage with Graceful Performance Degradation ANXIAO (ANDREW) JIANG California Institute of Technology and JEHOSHUA BRUCK California Institute of Technology A file storage scheme is proposed
More informationDiversity Coloring for Distributed Data Storage in Networks 1
Diversity Coloring for Distributed Data Storage in Networks 1 Anxiao (Andrew) Jiang and Jehoshua Bruck California Institute of Technology Pasadena, CA 9115, U.S.A. {jax, bruck}@paradise.caltech.edu Abstract
More informationLatency on a Switched Ethernet Network
Application Note 8 Latency on a Switched Ethernet Network Introduction: This document serves to explain the sources of latency on a switched Ethernet network and describe how to calculate cumulative latency
More informationSocial Media Mining. Graph Essentials
Graph Essentials Graph Basics Measures Graph and Essentials Metrics 2 2 Nodes and Edges A network is a graph nodes, actors, or vertices (plural of vertex) Connections, edges or ties Edge Node Measures
More informationOptimizing the Placement of Integration Points in Multihop Wireless Networks
Optimizing the Placement of Integration Points in Multihop Wireless Networks Ranveer Chandra, Lili Qiu, Kamal Jain, Mohammad Mahdian Cornell University Microsoft Research Abstract Efficient integration
More informationLoad Balancing Mechanisms in Data Center Networks
Load Balancing Mechanisms in Data Center Networks Santosh Mahapatra Xin Yuan Department of Computer Science, Florida State University, Tallahassee, FL 33 {mahapatr,xyuan}@cs.fsu.edu Abstract We consider
More informationA simpler and better derandomization of an approximation algorithm for Single Source RentorBuy
A simpler and better derandomization of an approximation algorithm for Single Source RentorBuy David P. Williamson Anke van Zuylen School of Operations Research and Industrial Engineering, Cornell University,
More informationFactoring & Primality
Factoring & Primality Lecturer: Dimitris Papadopoulos In this lecture we will discuss the problem of integer factorization and primality testing, two problems that have been the focus of a great amount
More informationDistributed Computing over Communication Networks: Maximal Independent Set
Distributed Computing over Communication Networks: Maximal Independent Set What is a MIS? MIS An independent set (IS) of an undirected graph is a subset U of nodes such that no two nodes in U are adjacent.
More informationAn Efficient Hybrid Data Gathering Scheme in Wireless Sensor Networks
An Efficient Hybrid Data Gathering Scheme in Wireless Sensor Networks Ayon Chakraborty 1, Swarup Kumar Mitra 2, and M.K. Naskar 3 1 Department of CSE, Jadavpur University, Kolkata, India 2 Department of
More informationCost Model: Work, Span and Parallelism. 1 The RAM model for sequential computation:
CSE341T 08/31/2015 Lecture 3 Cost Model: Work, Span and Parallelism In this lecture, we will look at how one analyze a parallel program written using Cilk Plus. When we analyze the cost of an algorithm
More informationPolicyBased Fault Management for Integrating IP over Optical Networks
PolicyBased Fault Management for Integrating IP over Optical Networks Cláudio Carvalho 1, Edmundo Madeira 1, Fábio Verdi 2, and Maurício Magalhães 2 1 Institute of Computing (ICUNICAMP) 13084971 Campinas,
More information5.1 Bipartite Matching
CS787: Advanced Algorithms Lecture 5: Applications of Network Flow In the last lecture, we looked at the problem of finding the maximum flow in a graph, and how it can be efficiently solved using the FordFulkerson
More informationHow To Provide Qos Based Routing In The Internet
CHAPTER 2 QoS ROUTING AND ITS ROLE IN QOS PARADIGM 22 QoS ROUTING AND ITS ROLE IN QOS PARADIGM 2.1 INTRODUCTION As the main emphasis of the present research work is on achieving QoS in routing, hence this
More informationDiscrete Mathematics & Mathematical Reasoning Chapter 10: Graphs
Discrete Mathematics & Mathematical Reasoning Chapter 10: Graphs Kousha Etessami U. of Edinburgh, UK Kousha Etessami (U. of Edinburgh, UK) Discrete Mathematics (Chapter 6) 1 / 13 Overview Graphs and Graph
More informationOffline sorting buffers on Line
Offline sorting buffers on Line Rohit Khandekar 1 and Vinayaka Pandit 2 1 University of Waterloo, ON, Canada. email: rkhandekar@gmail.com 2 IBM India Research Lab, New Delhi. email: pvinayak@in.ibm.com
More informationAn Algorithm for Automatic Base Station Placement in Cellular Network Deployment
An Algorithm for Automatic Base Station Placement in Cellular Network Deployment István Törős and Péter Fazekas High Speed Networks Laboratory Dept. of Telecommunications, Budapest University of Technology
More informationInterconnection Networks. Interconnection Networks. Interconnection networks are used everywhere!
Interconnection Networks Interconnection Networks Interconnection networks are used everywhere! Supercomputers connecting the processors Routers connecting the ports can consider a router as a parallel
More informationPh.D. Thesis. Judit NagyGyörgy. Supervisor: Péter Hajnal Associate Professor
Online algorithms for combinatorial problems Ph.D. Thesis by Judit NagyGyörgy Supervisor: Péter Hajnal Associate Professor Doctoral School in Mathematics and Computer Science University of Szeged Bolyai
More informationIntroduction to LAN/WAN. Network Layer
Introduction to LAN/WAN Network Layer Topics Introduction (55.1) Routing (5.2) (The core) Internetworking (5.5) Congestion Control (5.3) Network Layer Design Isues StoreandForward Packet Switching Services
More informationDynamic Establishment of AllOptical Connections in WavelengthRouted Optical Networks Using Genetic Algorithms
D. Bisbal, F. González, I. de Miguel, J. Blas, J. C. Aguado, P. Fernández, R. M. Lorenzo, E. J. Abril, M. López, Dynamic establishment of alloptical connections in wavelengthrouted optical networks using
More informationPerformance advantages of resource sharing in polymorphic optical networks
R. J. Durán, I. de Miguel, N. Merayo, P. Fernández, R. M. Lorenzo, E. J. Abril, I. Tafur Monroy, Performance advantages of resource sharing in polymorphic optical networks, Proc. of the 0th European Conference
More informationA Game Theoretical Framework on Intrusion Detection in Heterogeneous Networks Lin Chen, Member, IEEE, and Jean Leneutre
IEEE TRANSACTIONS ON INFORMATION FORENSICS AND SECURITY, VOL 4, NO 2, JUNE 2009 165 A Game Theoretical Framework on Intrusion Detection in Heterogeneous Networks Lin Chen, Member, IEEE, and Jean Leneutre
More informationAlgorithm Design for Performance Aware VM Consolidation
Algorithm Design for Performance Aware VM Consolidation Alan Roytman University of California, Los Angeles Sriram Govindan Microsoft Corporation Jie Liu Microsoft Research Aman Kansal Microsoft Research
More informationNEW applications of wireless multihop networks, such
870 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 17, NO. 3, JUNE 2009 Delay Aware Link Scheduling for MultiHop TDMA Wireless Networks Petar Djukic, Member, IEEE, and Shahrokh Valaee, Senior Member, IEEE
More information