International Journal of Electronics and Communications (AEÜ)

Transcription

1 Int. J. Eectron. Commun. (AEÜ) 67 (2013) Contents ists avaiabe at SciVerse ScienceDirect Internationa Journa of Eectronics and Communications (AEÜ) journa h o me pa ge: Hybrid FRR/p-cyce design for ink and node protection in MPLS networks Chang Cao a,b,, George N. Rouskas c, Jianquan Wang d, Xiongyan Tang d a China United Network Communications Co., Ltd., Postdoctora Workstation, Beijing , China b Beijing University of Posts and Teecommunications, Beijing , China c Department of Computer Science, North Caroina State University, Raeigh, NC , USA d China Unicom Research Institute, Beijing , China a r t i c e i n f o Artice history: Received 7 Juy 2012 Accepted 11 November 2012 Keywords: Muti-protoco abe switching Fast reroute Pre-configure cyce a b s t r a c t Survivabe MPLS technoogies are crucia in ensuring reiabe communication services. The fast reroute (FRR) mechanism has been standardized to achieve fast oca repair of abe switched paths (LSPs) in the event of ink or node faiures. We present a suite of hybrid protection schemes for MPLS networks that combine the we-known p-cyce method with FRR technoogy. Whereas with pure FRR backup paths are panned by each node individuay, the hybrid schemes empoy a set of p-cyces that may be seected using techniques that take a hoistic view of the network so as to share protection bandwidth effectivey. The hybrid FRR/p-cyce methods are fuy RFC 4090-compiant, yet aow network operators to everage a arge existing body of p-cyce design techniques. Numerica resuts on reaistic network topoogies indicate that the hybrid approach is successfu in combining the advantages of p-cyce design and FRR Esevier GmbH. A rights reserved. 1. Introduction Muti-protoco abe switching (MPLS) [1], originay deveoped to enabe fast packet forwarding, has aso faciitated traffic engineering, quaity-of-service (QoS) routing, and differentiated services support in IP-based metro and backbone networks [2]. MPLS technoogy is widey depoyed and is crucia to the operation of the Internet and its abiity to support critica communication services efficienty. Consequenty, MPLS survivabiity mechanisms [3] are key to ensuring that the network may continue to provide reiabe services even in the presence of faiures. In particuar, with today s muti-ayer network architectures, it may be more economica for IP/MPLS ayer operators to restore traffic within their own IP/MPLS ogica environment rather than reying on physica ayer restorabiity [4 6]. There are mainy two types of faiures that network operators must design the network to withstand. Link faiures (e.g., due to fiber cuts or the mafunction of active components such as transponders) are usuay handed by the physica ayer first. But if such a faiure is not restored within a certain period of time (typicay, a few tens of miiseconds), the MPLS ayer must initiate its own recovery actions. Node faiures may be due to router crashes Chang Cao was supported by China Nationa Science and Technoogy Major Project (2010ZX ). Corresponding author at: China United Network Communications Co., Ltd., Postdoctora Workstation, Beijing , China. Te.: E-mai address: caochang@chinaunicom.cn (C. Cao). or router restarts after the routine appication of software patches or upgrades, and may occur as frequenty as, or even more often than ink faiures [7]. Such faiures must be deat with directy at the MPLS ayer. The IETF has standardized the fast reroute (FRR) mechanism [8] for protecting abe switched paths (LSPs) in MPLS networks. FRR cas for oca repair actions in the event of a ink or node faiure. Specificay, affected traffic is re-directed onto pre-configured backup tunnes by two nodes adjacent to the faied ink or node. As a resut, a affected LSPs are rerouted to backup paths within a few tens of miiseconds. The pre-configure cyce (p-cyce) scheme [9] aso empoys oca repair actions to re-direct traffic from the faied ink or node onto a backup path aong a pre-configured cyce. This method provides ring-ike protection speeds with mesh (span-restorabe) capacity efficiency, and it has been studied extensivey (for a survey of reated work, see [10]). Athough originay designed for protection in the optica ayer, p-cyce technoogy can be appied to the IP [7] or MPLS [11] ayers. There is an extensive technica iterature on network survivabiity design, optimization, and performance evauation using p-cyces in the WDM and IP ayers [10]. For packet-switched networks, in particuar, it was shown in [7] that by using integer inear programming (ILP) methods, p-cyce design can be as capacityefficient as optimized span restoration. However, obtaining exact optima soutions is an NP-hard combinatoria probem that does not scae to arge networks. As a resut, a number of reaxations and heuristics must be considered for practica appication of p-cyce seection to reaistic network topoogies. Another two studies /$ see front matter 2012 Esevier GmbH. A rights reserved.

2 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) (first [11] and then [12]) presented a group of mixed integer program (MIP) formuations for p-cyce design in MPLS networks, and investigated the reationship between protection bandwidth requirements and traffic oad distribution. Besides MIP formuations discussion, study in [13] focused on the issue of p-cyces protection switching protocos. It expained that p-cyces may not be abe to recover a traffic transiting through a faied node as rings did, and proposed a protoco enhancement which protected quaified paths against node faiures. The concept of path-segment protecting p-cyces was first described in [14], it extended the abiity of p-cyces to protect node by restoring a of reevant path segments. Later, this method was deveoped in [15], which et the cyces act as p-cyces for end-to-end paths between nodes on the cyce, but ony aowed each cyce to provide protection reationships to a group of paths whose routes are a mutuay disjoint. In order to simpify the design for node protection, study in [16] reported a new strategy that integrated the BLSR-ike behavior of ordinary p-cyces under on-cyce node faiure conditions with a new stradding-oriented use of the p-cyces for node protection, and empoys ony one set of p-cyces over the whoe network. More recenty, a new shared-segment protection to restore node faiure using ordinary p-cyces was introduced in [17]. The motivation for our work is based on the observation that both FRR and p-cyce are oca repair protection schemes, hence we expect the network operation, deay, and overhead incurred for faiure detection, notification, and triggering of restoration action to be simiar for the two technoogies. However, FRR backup tunnes are typicay panned individuay by the nodes adjacent to the protected ink or node, whereas p-cyce design takes a more hoistic view of the network in determining the protection cyces so as to share spare resources effectivey. Therefore, we first introduce the signaing and oca repair (protection switching) methods of FRR into p-cyce s restoration, which are not empoyed in p-cyces operation before, and then propose severa nove design technique (i.e., area division p-cyces and FRR-based p-cyces). Together with panning methods of Hamitonian p-cyce and node encircing p-cyce in previous studies [10], we compare these four types of p-cyces performance in MPLS networks with different topoogies. Another contribution of our work is to introduce network hoistic p-cyce designs, which means to use the same set of p-cyces to protect both ink and node faiures. The rest of the paper is organized as foows. In Section 2, we briefy review the FRR method for MPLS ink and node protection. In Section 3, we describe how to combine the p-cyce and FRR methods, and describe severa approaches for seecting the set of p-cyces. In Section 4 we present three performance metrics to compare the pure FRR and hybrid schemes. We present numerica resuts in Section 5, and concude the paper in Section Pure FRR protection design RFC 4090 [8] defines two schemes for deaing with ink and node faiures, respectivey. Consider first the case of ink faiure, e.g., the faiure of directed ink B C in the 8-node MPLS network shown in Fig. 1. The upstream router B is referred to as the point of oca repair (PLR) with respect to protecting traffic in the event that the ink fais, whie the downstream router C adjacent to the other end of the ink is known as the merge point (MP). On the other hand, if a node fais, a inks incident to the node (i.e., the three bidirectiona inks C D, D H, and D G, in the case of faied node D in the network of Fig. 1) are affected. In other words, we can regard a node faiure as the simutaneous oss of a ink pairs (i.e., 2-hop paths) with the faied node in the midde. In this situation, a neighbors of the faied node act as the PLR or MP, where the specific Fig. 1. Link/node protection with pure FRR. roe of each neighbor is determined by the direction of the specific traffic fow considered. Regardess of whether the faiure invoves a ink or node, impementation of FRR protection consists of three steps [8]: 1. Panning. The key idea in FRR is to find, for each protected ink or ink pair and before any faiure takes pace, a backup path from the PLR node to the MP node. Consider first the case of ink faiure, e.g., of directed ink B C in Fig. 1. In MPLS FRR, the MP node C is the next-hop (NHop) router of the PLR node B with respect to the ink B C. Hence, B may seect the path B E F C on which to re-direct traffic in the event that ink B C fais. Let us now consider the case of node faiure, e.g., of node D in Fig. 1. Suppose that node C, a neighbor of D, has traffic that passes through D on its way to node H and beyond, i.e., traverses the ink pair (C D, D H). In FRR, from the point of view of protecting this ink pair, C is the PLR node and H is the next-nexthop (NNHop) MP node. The NNHop scheme consists of finding a backup path from the PLR to the NNHop router; Fig. 1 shows that the path C F G H has been seected to protect the ink pair (C D, D H). Simiar actions are taken by a neighbors of the faied node D to protect a ink pairs through this node. 2. Backup LSP signaing. Backup LSPs are estabished aong the backup paths using the same signaing mechanisms (e.g., RSVP- TE) as for setting up working LSPs. Athough backup LSPs do not carry traffic under norma conditions, they are ready to accept traffic re-directed from faied working LSPs once the backup signaing is finished. This step is impemented identicay for both ink and node protection. 3. Loca repair. When a ink (e.g., ink B C in Fig. 1) or a node (e.g., node D in Fig. 1) fais, the faiure wi be detected and confirmed after severa signaing actions between its PLR and MP nodes. From that instant, and unti a goba routing update takes effect, any packets that woud have been forwarded aong the faied ink or node, are instead re-directed by the PLR node onto the corresponding backup path, e.g., as shown in Fig. 1. In re-directing traffic affected by the faiure, the PLR uses a new abe (i.e., FRR, or protection, abe) in pace of the former working abe, so that packets be forwarded aong the backup LSP. Once the packets arrive at the MP node over the backup LSP, they are forwarded toward their destination as if they had arrived over the working LSP. Note that the second (signaing) and third (oca repair) steps of this process must conform to the reevant MPLS standards, especiay RFC 4090 [8]. However, the first step (panning) is outside the purview of standards, and network operators are free to empoy customized agorithms to seect a backup path for each protected ink. In pure FRR, the backup path for each protected ink or ink pair (in case of node faiure) is seected by the PLR node, typicay using a constraint-based shortest path first (CSPF) agorithm [8, Section 6.2]. Since the PLR of a protected ink/ink pair executes the CSPF agorithm independenty of other routers, it makes a ocay optima decision based on its own information.

3 472 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) Fig. 2. Hybrid FRR/p-cyce protection. However, these ocay optima backup paths may not constitute a gobay optima soution, i.e., one that optimizes a network-wide objective such as backup resource cost and/or utiization. Since the panning step may take pace offine, it is possibe to empoy a more sophisticated design methodoogy that takes a more hoistic view of network performance and cost in seecting backup paths. In the next subsection we wi describe how to appy such a backup capacity design based on the p-cyce concept. 3. Hybrid FRR/p-cyce design The p-cyce design is simpy an aternative way of carrying out the panning step of ink or node protection. In this step, the whoe network may be protected by a singe Hamitonian p-cyce or a set of smaer p-cyces that may be seected according to various methodoogies. As we mentioned in Section 1, there has been extensive research in deveoping optimization techniques for seecting optima sets of p-cyces. However, genera variants of the probem are NP-hard, with the computationa compexity increasing quicky with the size and density of the network. As reported in [18], hopimited p-cyce designs may take severa hours to sove optimay even for a reativey sma network. Our goa in this work is not to present optima p-cyce soutions, but rather, to quantify the benefits of incorporating such designs within the FRR framework. To this end, we consider four schemes for determining p-cyce sets that can be used to protect an MPLS network from a singe ink or node faiure. Each method uses a fast technique to yied a good p-cyce set; taken as a whoe these p-cyce sets are representative of the performance improvements that are achievabe reative to pure FRR. To the degree that optima p-cyce techniques might yied additiona performance improvements, they woud provide further support to our argument of using a hybrid FRR/p-cyce design. Nevertheess, quantifying the benefits of optima design is outside the scope of this work. We now discuss four schemes for seecting p-cyces, as iustrated in Fig Hamitonian p-cyce. The concept of Hamitonian p-cyce has been known for years, and it generay works we for sma size networks. As shown in Fig. 2, a singe Hamitonian p-cyce may be configured through a eight nodes of the network. Such a p-cyce may restore any singe ink or node faiure by having the PLR reroute the affected traffic toward the MP router, aong the part of the cyce that is accessibe to the PLR after the faiure. Whie a Hamitonian represents a straightforward p-cyce soution for protection, the restoration time increases with the size of the network due to the ong backup paths. 2. Area division p-cyces (ADPC). Rather than using a singe Hamitonian p-cyce for the whoe network, the key idea is to divide the network into severa smaer areas and buid a Hamitonian cyce for each area; the resuting set of p-cyces is then used for ink and/or node protection in the origina network. Therefore, the areas must be determined such that coectivey, they incude a network inks (for ink protection ony), a ink pairs (for node protection ony), or both (for ink and node protection). By carefuy seecting the areas, the ength of the corresponding Hamitonian cyces (and, hence, of the backup paths) can be kept we beow that of a singe Hamitonian p-cyce, without sacrificing backup resources (as the numerica resuts, to be presented shorty, indicate). 3. Node encircing p-cyces (NEPC). Node encircing p-cyces (NEPC) have been proposed for node protection at the IP/MPLS ayer [7]. Each such cyce is designed to protect a specific network node, and incudes a the neighbors of this node (but not the node itsef). We observe that by seecting NEPCs appropriatey, a inks of the network may be covered as we. Therefore a set of NEPCs may aso be used for ink protection. 4. FRR-based p-cyces (FRRPC). Observe that, taken together, the working and protection paths used by pure FRR to protect from a given ink or ink pair faiure form a cyce. Therefore, we may buid a set of p-cyces based on the working and backup paths constructed by FRR, referred to as FRR-based p-cyces (FRRPC). The main difference from pure FRR is that by organizing the paths into p-cyces, backup capacity may be shared among backup paths that protect different inks and/or nodes. This is simpy not possibe for pure FRR since each node individuay constructs backup tunnes to protect its adjacent inks or nodes, hence there is no capacity sharing among the backup paths constructed by different nodes. Since this set of p-cyces uses the same routing as pure FRR, it aows us to investigate the potentia improvement in backup capacity requirements that is due to simpy taking a comprehensive view of protecting the whoe network through p-cyces, rather than having each node make its own protection arrangements independenty of other nodes in the network. Note that, in the context of MPLS networks, a p-cyce is a ogica entity, and its purpose is simpy to define the backup path for each ink or node that it protects. Consider the ink B C in Fig. 2 which happens to be an on-cyce ink for the Hamitonian and ADPC p- cyces shown. The backup path for this ink is the path from B to C aong the counter-cockwise direction on either p-cyce. Simiary, a on-cyce inks are backed up by the (unique) reverse path to their MP node. On the other hand, there are two potentia backup paths for each stradding ink (e.g., ink B E with respect to the Hamitonian), one aong each direction around the p-cyce. Operators may use one of these two paths (e.g., the shortest one), or both. Simiar observations hod for node protection. Once the set of p-cyces has been seected, the second (signaing of backup LSPs) and third (oca repair) steps take pace exacty as the standard [8] specifies. We discuss the oca repair step in more detai in Section Performance metrics and anaysis Now we discuss three performance metrics to evauate the reative merits of the pure FRR and hybrid FRR/p-cyce schemes. Our goa is to protect the network from any one of three scenarios: (1) singe ink faiure ony, (2) singe node faiure ony, or (3) either a singe ink or singe node faiure. For the pure FRR scheme, we assume that the backup path of each ink or node is given; whie for the hybrid scheme we assume that the p-cyce set is given and that each stradding ink (in a ink faiure case) or ink pair (in a node faiure case) is protected by sending its traffic aong the two backup paths around the p-cyce. Uness we expicity specify otherwise, whenever we refer to a ink we assume that the ink is directed. Symbos that are used in our metrics are defined in Tabe 1.

4 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) Tabe 1 Symbos used for performance metrics. Symbos Meanings Symbos Meanings B Tota backup capacity W Tota working capacity W Working capacity carried by any ink W 1, 2 Working traffic that fows on both inks of pair ( 1, 2) W s Working capacity on the short backup path of a cyce W Working capacity on the ong backup path of a cyce p Backup path for ink protection p 1, 2 Backup path for ink pair ( 1, 2) protection d Length of backup path for ink protection d 1, 2 Length of backup path for ink pair ( 1, 2) protection L in n /Lout n Set of incoming/outgoing inks of node n L cw c /L ccw c Set of cockwise/counter-cockwise inks of p-cyce c L str c Set of stradding inks of p-cyce c B frr Tota amount of backup capacity for pure FRR H n Weighted backup hop cost for node n B hfrr Tota amount of backup capacity for hybrid FRR/p-cyce B Backup capacity for on-cyce inks if ink fais B on /B off Backup capacity for on-cyce/stradding ( 1, 2) protection d c Length of a p-cyce c dc s /d c Length of the short/ong backup path on a p-cyce c H frr Weighted backup hop cost for pure FRR H hfrr Weighted backup hop cost for hybrid FRR/p-cyce H Weighted backup hop cost for ink H 1, 2 Weighted backup hop cost for ink pair ( 1, 2) In order to give better anaysis, we put working capacity with different units on spans of Fig. 3. Furthermore, here we assume that ony foowing 8 inks are needed to be protected, which are in cockwise direction as A B C D H G F E A. Working capacity of these directed inks are set as 3, 5, 1, 3, 3, 2, 5, 4 respectivey. The backup paths for traffic restoration compies with the design of either FRR or hybrid FRR/p-cyce schemes, and here the panning of ADPC is indicated with three cyces of different coors B/W ratio By setting up bandwidth-guaranteed backup LSPs, it is possibe for the MPLS network to protect a working LSPs. The B/W ratio is an important metric to compare p-cyces performance, which has been used in ots of iterature ike [7 15]. Given the routing and amount of traffic carried by each working LSP, it is straightforward to compute the tota working traffic W carried by any ink (assuming the routing of working LSPs is independent of how backup LSPs are seected). Therefore, W in the network is equa to: W = W. Assuming that in each of the three scenarios we consider, the network must be protected from a ink and/or a node faiures, then the amount of working capacity to be protected is equa to W in each case. In the foowing subsections, we derive expressions for the backup capacity B that is needed for each scenario under the pure FRR and hybrid FRR/p-cyce schemes Backup capacity for pure FRR Let us first consider protection from singe ink faiure ony (first scenario). In this case, an amount of backup capacity equa to W must be provisioned aong each ink of the backup path for ink. Let p be the backup path for ink, and et d be the ength (in hops) of path p. Then, the tota amount of backup capacity for FRR under scenario 1 is: L B 1 frr = d W, (1) =1 Fig. 3. Demonstration of spans working capacity and ADPC panning. where L is the number of (directed) inks in the network. For exampe, in order to protect 8 seected (directed) inks in Fig. 3, the tota aocated backup capacity wi be cacuated as: =65 Consider now scenario 2, i.e., protection from singe node faiures ony. Whenever a node n fais, a the traffic on the inks adjacent to n (in either direction) must be protected, except traffic that originates or terminates at node n. Let L in n and L out n denote the set of incoming and outgoing inks, respectivey, of node n. Let ( 1, 2 ) be a pair of (directed) inks passing through n, i.e., 1 L in n and 2 L out n. Let p 1, 2 be the backup path that is used to bypass this ink pair in the event that n fais, and d 1, 2 be the ength of this path. Let W 1, 2 denotes the amount of working traffic that fows on both inks of pair ( 1, 2 ), and must be protected if n fais; in other words, this is the traffic that traves from ink 1 to ink 2 through node n, not incuding any traffic that terminates at, or originates from, n. The tota amount of backup capacity required for scenario 2 can then be obtained as: B 2 frr = N n=1 1 L in n, 2 Lout n d 1, 2 W 1, 2 (2) where N is the tota number of nodes in the network. For instance, if the working capacity of ink pair A B C is 3, and E B C is 2, totay =13 unit wi be cost to protect node B. As for other nodes, the principe is simiar. For scenario 3 (i.e., protection from either a singe ink or a singe node faiure), both the NHop and NNHop schemes must be activated independenty under pure FRR. Hence, the tota amount of protection bandwidth required in this case is: B 3 frr = B 1 frr + B 2 frr. (3) Backup capacity for hybrid FRR/p-cyce For ink protection, et us assume that C, C 1, p-cyces have been configured for protecting the network inks. Note that, if a ink is a stradding ink in some p-cyce c, then an amount of backup capacity equa to W /2 on the on-cyce inks (in both directions) is sufficient to protect this ink. On the other hand, if a ink is an oncyce ink, then W units of backup capacity need to be provisioned on a other inks of the p-cyce in the opposite direction. However, if a ink is an on-cyce ink of k different p-cyces, each p-cyce needs to provision ony W /k units of backup capacity. Let L cw c (respectivey, L ccw c ) denotes the set of cockwise (respectivey, counter-cockwise) inks of p-cyce c, and L str c denote the set of stradding inks of p- cyce c. Based on these observations, for scenario 1, protection from

5 474 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) a singe ink faiure ony, the backup capacity for the cockwise oncyce inks of p-cyce c is given by: { { B 1 W } { W } } = max max, max, L cw L ccw k c L str c. (4) 2 c A simiar expression can be written for the backup capacity of counter-cockwise inks, whie stradding inks need no backup capacity, i.e., B 1 = 0, L str. We aso note that if a ink beongs to mutipe p-cyces, the backup capacity that is reserved on this ink is the maximum capacity assigned by any p-cyce c from the corresponding expression (4). According to the Fig. 3, since the biggest vaue of each ink s working capacity is 5, for a Hamitonian p-cyce, it ony costs 8 5 =40 unit to protect any singe ink faiure. If we use ADPC or other p-cyce design, by putting area division ike Fig. 3 and cacuating the protection capacity in each smaer Hamitonian p-cyce, =50 wi be used for the whoe ink protection. For scenario 2 (protection from singe node faiure ony), we distinguish between two cases. If a node n on a p-cyce fais, the p-cyce can be used to to protect a ink pair ( 1, 2 ) through node n if (1) both inks are on-cyce inks, (2) one ink is an on-cyce ink and the other one is a stradding ink on the p-cyce, or (3) both inks are stradding inks of the p-cyce. In each of these cases, there is ony one backup path around the p-cyce to restore affected traffic, and the amount of capacity on each ink of the backup path is equa to: B on = max{w 1, 2 }, where the maximum operation is over a ink pairs ( 1, 2 ) for which on-cyce ink is on their backup path. For exampe, if node B faiure happens in Fig. 3, ink pair A B C wi be protected by the other part of a Hamitonian p-cyce going from A through C, assuming traffic on A B C is 3 unit, it wi cost 6 3 =18 unit for this ink pair protection. However, as for ADPC, it ony uses 3 3 =9 unit to restore the faiure. If a node n not on the p-cyce fais, the p-cyce can be used to protect a ink pair ( 1, 2 ) if the path ( 1, 2 ) is a stradding path of the p-cyce (e.g., as in node-encircing p-cyces). In this case, an amount of capacity equa to B off = max{w 1, 2 /2} must be reserved on each on-cyce ink, where the maximum operation is taken over a ink pairs ( 1, 2 ) that are stradding paths of the p-cyce. Hence, under scenario 2, the backup capacity for the on-cyce inks of p-cyce c is given by: { } B 2 = max B on, B off, L cw c L ccw c. (5) As in scenario 1, stradding inks need no backup capacity. For scenario 3, i.e., protection from either singe ink or singe node faiure, the backup capacity on each ink must be equa to the maximum of the capacities required for scenarios 1 and 2, i.e., B 3 = max{b 1, B 2 }. (6) Consequenty, the tota backup capacity for hybrid FRR/p-cyce under scenario i, i = 1, 2, 3, can be computed as: L B i hfrr = B i, i = 1, 2, 3. (7) = Traffic weighted backup hop cost When a ink or a node fais, a traffic on the affected ink(s) is re-directed aong the backup path(s), incurring additiona deay that depends on the ength of backup paths. Here traffic weighted backup hop cost is used for measuring that in order to protect one ink or node of the whoe network, how many additiona resources are needed in average. This is a new metric that first introduced by this paper Pure FRR Let d denotes the ength (in hops) of path p that serves as the backup path of ink. If ink, carrying an amount W of working traffic fais, the traffic weighted backup hop cost incurred by ink LSPs is given by: H 1 = W d. Assuming that a L inks are equay ikey to fai, the average traffic weighted backup hop cost for pure FRR under scenario 1 can be written as: L H 1 frr = =1 H1 (8) L Since we have worked out that it wi add totay 65 unit to protect a of singe ink faiure in Section 4.1.1, for every ink, the weighted backup hop cost is 65/8 = In scenario 2, whenever a node n fais, traffic on a ink pairs through this node is re-directed on the corresponding backup paths. Using our earier notation, the traffic weighted backup hop cost for node n is: Hn 2 = W 1, 2 d 1, 2. (9) 1 L in n, 2 Lout n Consequenty, the average cost over a node faiures is: N H 2 frr = n=1 H2 n (10) N As a resut, for pure FRR case, we can get the average cost over a node faiures by first adding the amount of every node faiure cost together, and then divide this vaue by the number of nodes. For scenario 3, if we assume that a ink or node faiures are equay ikey, then the traffic weighted backup cost can be given as: L H 3 frr = =1 H1 N + n=1 H2 n (11) L + N If node and/or ink faiures have different probabiity of occurring, the above expression can be modified in a straightforward manner. Since our objective is to investigate the reative costs of the pure and hybrid FRR schemes, we wi use the above expression for simpicity Hybrid FRR/p-cyce For the FRR/p-cyce hybrid scheme, again assume that C p-cyces have been configured, and et d c 3 denote the ength (i.e., number of directed on-cyce inks) of p-cyce c, c = 1,..., C. Consider, first, scenario 1. If ink is an on-cyce ink for k p-cyces, the traffic weighted backup hop cost for this ink is: k H 1 W = (d c 1). (12) k j=1 For a ink that is a stradding ink on p-cyce c, et dc s and dc denote the ength of the short and ong backup paths, respectivey, for the ink aong the p-cyce, i.e., such that dc s dc and dc s + dc = d c. We send as much working traffic W s as possibe on the short backup path, i.e., W s = min{w, B c }, where B c is the spare capacity on the on-cyce inks of the p-cyce, and the remaining traffic, W = W W s, if any, on the ong backup path. Hence, the weighted cost is: H 1 = W s d s c + W d c. (13) The average traffic weighted backup cost, H 1 hfrr, can be obtained by an expression simiar to (8). According to the Fig. 3, since 3 unit is needed to go through a other spans to protect ink A B, 5 unit to protect ink B C, simiary, totay ( ) (8 1) = 189 unit wi be reserved for every ink faiure. Consequenty, the weighted backup cost is 189/8 =

6 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) As for ADPC scheme, it wi first cacuate three cyces as: ( ) (5 1) + ( ) (5 1) + ( ) (6 1) = So the weighted backup cost is 110.4/8 =13.8. For scenario 2, if the ink pair ( 1, 2 ) faied due to the faiure of a node on a p-cyce, then H 1, 2 = W 1, 2 d 1, 2, (14) where d 1, 2 is the ength of the backup path for this ink pair on the p-cyce. If the ink pair ( 1, 2 ) is a stradding path in a p-cyce, then H 1, 2 = W s 1, 2 d s c + W 1, 2 d c. (15) The cacuation is quite ike we have demonstrated for node B with pure FRR scheme. The average traffic weighted backup cost, H 2 hfrr, can be obtained by expressions simiar to (9) and (10). Finay, the average traffic weighted backup hop cost for scenario 3, H 3 hfrr, can aso be obtained by using an expression simiar to (11) Labe entry overhead The number of abes required to estabish backup paths is an important metric for MPLS networks, as it determines the size of the forwarding tabes at the LSRs. The metric of abe entry overhead has been mentioned in some references ike [1,2], and here it is first empoyed to compare the performance of MPLS protection. We assume that the one-to-one backup method [8] is used to impement the oca repair technique. This method requires the aocation of a different set of protection abes for every traffic component (i.e., LSP). For exampe, if ink fais, a traffic on the ink is sent aong the backup path by having the PLR node switch on each affected packet the protection abe associated with the LSP of the packet. Each intermediate node on the backup path forwards packets based on their protection abe, and repaces it with a new protection abe, as per the norma MPLS packet forwarding operation. When the MP node receives a packet with a protection abe, it repaces it with a new working abe and forwards the packet aong its origina (working) path. In case of node faiure, the operation for every node aong the backup path (incuding the PLR and MP nodes) is identica. Therefore, the number of additiona abes required to protect a ink or a ink pair ( 1, 2 ) with the one-to-one backup method is equa to the number of hops aong the corresponding backup path times the number of traffic components (LSPs) that traverse this ink or ink pair, respectivey. Since the backup paths in the hybrid FRR/p-cyce scheme are embedded into p-cyces that are determined in advance, it is possibe to use a smaer number of protection abes. Consider first the case of ink faiure, and observe that the backup path aways traverses a the inks of the p-cyce (in the opposite direction of the working ink that faied). Hence, we assign ony two sets of protection abes for each p-cyce, one set in each direction (cockwise or counter-cockwise). If a ink fais, for each affected packet, the PLR node (1) switches the incoming working abe to the appropriate outgoing working abe (as in norma operation), (2) pushes the same outer protection abe onto a packets, and (3) forwards a such traffic aong the backup path on the appropriate p-cyce. Intermediate nodes on the backup path simpy switch the appropriate protection abe assigned for the p-cyce and direction of the backup path. When the MP node at the other end of the backup path receives a packet with a protection abe, it pops this abe and continues to forward the packet based on the inner working abe assigned by the PLR node. Therefore, for ink protection, the number of abes required for a the backup paths on a p-cyce is simpy twice the number of inks in the p-cyce; i.e., one set of abes for each direction aong the p-cyce. This arrangement is possibe because each node on the p-cyce may reuse the same set of abes to accommodate any (on-cyce or stradding) ink faiure without ambiguity: under any faiure, ony the MP node of the faied ink is aware of the faiure and is the one to remove traffic redirected due to the faiure on the bypass tunne from the p-cyce. In the case of node faiure, backup paths for the various protected ink pairs aso foow a p-cyce. However, backup paths for different ink pairs affected by the faiure may terminate at different MP nodes; more importanty, even if backup paths terminate at the same MP node, the traffic on these backup paths may have to take different routes after reaching the MP node, depending on the specific ink pair that each backup path protects. Therefore, in addition to the two sets of protection abes that are associated with each direction of the p-cyce (as in the ink faiure case), we introduce one additiona protection abe for each ink pair protected by the p-cyce. When a node fais, for each affected packet, the PLR node: (1) switches the incoming working abe to a protection abe associated with the protected ink pair that the packet woud have traversed under norma operation, (2) pushes the same outer protection abe onto a packets, and (3) forwards a such traffic aong the backup path aong the p-cyce. Intermediate nodes simpy switch the appropriate outer protection abe, forwarding the packet aong the p-cyce toward the MP node. When the MP node receives a packet with a protection abe, it pops this abe and examines the inner abe. If the inner abe is aso a protection abe (which necessariy corresponds to a specific ink pair), it forwards the packet onto the appropriate working path after first switching this inner abe with the corresponding working abe. If the inner abe is not a protection abe, then it must be a working abe corresponding to a ink faiure, and the MP node proceeds as we described above. With this arrangement, the number of abes required for each p-cyce is twice the ength of the cyce (as in the ink faiure case), pus the number, say, K of ink pairs protected by the p-cyce. 5. Numerica resuts We compare the pure FRR to the hybrid FRR/p-cyce schemes on a simuation testbed impemented using the OPNET modeer. For this performance study, we consider the three rea network topoogies shown in Fig. 4 that have been widey used in survivabiity research [18,19]. The Cost-239 (N = 11 nodes, L = 52 directed inks) topoogy iustrates a reativey dense network connecting 11 main cities in Europe, with an average node in-/out-degree D = 4.73, whie the Havana topoogy (N = 17, L = 52) demonstrates a reativey sparse network depoyed in Germany, with D = 3.06); The USA-20 topoogy (N = 20 nodes, L = 92 directed inks), shown in the right part of the figure, and has an average node in-/out-degree D = Note that here we ony evauate the performance of panning by empoying different virtua network design schemes, which can be simuated with a certain number of unit as traffic demands or span bandwidth. As for the physica networks, other two steps ike backup LSP signaing and oca repair are aso very important. They are suggested to compy with protocos of RSVP-TE [20] and RFC 4090 [8], and each of MPLS router in the network must impement these protocos. It wi usuay cost hundreds of miisecond to set up a new protection path. Traffic demands are set up between every pair of nodes in each network, and working traffic is routed aong shortest paths computed using Dijkstra s agorithm. Let t sd denotes the amount of working traffic carried by the LSP from s to d. To investigate the sensitivity of the reative performance of the pure FRR and the four hybrid schemes, we generated working traffic demands that foow four different patterns: Equa (EQ): t sd = t = constant, (s, d).

7 476 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) Fig. 4. Network topoogies used in the performance study. Uniform (UF): t sd is uniformy distributed in the interva [0, 20], (s, d). Locaity (LC): t sd is uniformy distributed in the interva [4(h h sd ), 4(h h sd + 1) 1], where h sd is the ength (in hops) of the shortest path between s and d and h is the ength of the ongest shortest path in the network; in this pattern, the traffic demand between each node pair decreases with the distance between the two nodes, and modes the traffic ocaity observed in some networks. Reverse ocaity (RL): t sd is uniformy distributed in the interva [4(h sd 1), 4h sd 1], where h sd is the ength (in hops) of the shortest path between s and d, hence, it increases with the ength of the shortest path h sd. For pure FRR, Dijkstra s agorithm was aso used to find the shortest backup path for each ink or ink pair ( 1, 2 ). For the hybrid FRR/p-cyce scheme, we use the four types of p-cyces that discussed in Section 3 for panning the backup paths. In particuar, reca that the ADPC scheme is based on subdividing each network into smaer areas, and seecting a Hamitonian cyce in each area. The various dotted ines in Fig. 4 indicate the smaer areas we used in each topoogy to seect p-cyces for the ADPC scheme. (a) Scenario 1: ink protection 5.1. Resuts and discussion The resuts of the simuation for the Cost-239, Havana and USA topoogies are presented in the two sub-figures of Figs. 5 and 6, respectivey. Each sub-figure corresponds to one of the three scenarios we consider, i.e., singe ink protection ony, singe node protection ony, or both singe ink and node protection. The subfigures pot the B/W ratio or the backup hop cost as a function of the traffic pattern (shown on the x axis). In order to make meaningfu comparisons, athough working traffic demands were generated according to the four traffic patterns described above, the tota working traffic in each case was set to 1000 units. In addition, these traffic patterns are potted with different suffixes as -C, -H and - U, which means the resuts are given for topoogies of Cost-239, Havana and USA respectivey. Each sub-figure contains five curves: one for the pure FRR scheme, and four for the hybrid FRR/p-cyce scheme corresponding to the four techniques for seecting p-cyces (as discussed in Section 3). Since Havana is quite a sparse topoogy, few simpe cyces coud be found around each node to form NEPCs. However, NEPCs with non-simpe cyces are not recommended for widey use, because it wi bring a ot of operation confusion and signaing compexity [16]. As a resut, here we ony empoy other three hybrid schemes for protecting the network of Havana. We first observe that under scenario 1 (ink protection ony), the costs associated with protection, i.e., the B/W ratio and backup hop cost, are higher than under scenario 2 (node protection ony). This resut can be expained by the fact that, when a ink fais, a the traffic on the ink must be re-directed onto backup paths. On the other hand, when a node fais, traffic originating or terminating at the node cannot be protected. Hence, over a possibe ink faiures, the amount of backup resources required is higher than over a (b) Scenario 2: node protection (c) Scenario 3: ink and node protection Fig. 5. Simuation resuts for the B/W ratio (a): scenario 1: ink protection; (b) scenario 2: node protection; (c) scenario 3: ink and node protection.

8 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) (a) Scenario 1: ink protection (b) Scenario 2: node protection be stradding spans on some p-cyces, hence increasing the protection efficiency (since stradding spans do not need to have any spare capacity). These resuts iustrate the benefits of taking a comprehensive view of the network and traffic demands in designing backup paths using p-cyces. Let us now turn our attention to the reative performance of the four hybrid schemes. The resuts indicate that a Hamitonian p- cyce is not aways the best option in terms of bandwidth efficiency. In fact, the ADPC, FRRPC and NEPC schemes tend to have better performance (even the owest B/W ratio)across a topoogies and the traffic patterns we considered in our study. In particuar, the NEPC scheme is aways the best for scenario 2, as it is designed specificay for protecting each node. Aso note that FRRPC uses the same paths as FRR (but arranged in p-cyces such that backup bandwidth is shared). The fact that FRRPC has generay a much ower B/W ratio than pure FRR is an indication of arranging backup paths so as to share protection resources. We aso note that on one hand, the traffic pattern does affect the B/W ratio, especiay for hybrid schemes, but the reative performance among the various schemes is simiar. Specificay, the ocaity (LC) pattern resuts in the owest amount of protection capacity: since the majority of traffic is between nodes cose to each other in distance, the corresponding backup paths are reativey short, resuting in ow overa spare capacity. Simiar arguments can be used to expain why the reverse ocaity (RL) pattern requires the highest amount of backup capacity among the four patterns considered here, whie the equa (EQ) and uniform (UF) patterns fa between the other two in terms of this metric. On the other hand, for the topoogy of Havana, the bandwidth resource for protection is even higher than the topoogy with bigger scae but more dense, i.e., the USA topoogy. That is because it is a sparse topoogy, ess spans can be used to protect mutipe inks/nodes, and more excusive bandwidth are reserved for each span to protect different inks/nodes. We further note that the various p-cyce sets we consider here were not optimized for any specific objective. Hence, the resuts of B/W ratio are ony an upper bound on what can be achieved using p- cyce design; using sophisticated optimization techniques to seect the p-cyce sets, additiona improvements in capacity efficiency woud be possibe Traffic weighted backup hop cost (c) Scenario 3: ink and node protection Fig. 6. Simuation resuts for the backup hop cost: (a) scenario 1: ink protection; (b) scenario 2: node protection; (c) scenario 3: ink and node protection. possibe node faiures. A simiar observation hods for the trafficweighted backup hop cost. Of course, under scenario 3 (protection of either singe ink or singe node faiures), the B/W ratio associated with protection is the highest among a scenarios B/W ratio From the two figures, it is evident that, under a three protection scenarios and a four traffic patterns, the B/W ratio for the pure FRR scheme is higher than that of the hybrid schemes. With pure FRR, there is no sharing of backup resources as its node pans its own backup paths independenty of other nodes. The hybrid designs, on the other hand, are abe to share protection bandwidth aong the p-cyces. Moreover, in reativey dense topoogies such as Cost-239 and USA, there are ampe opportunities for inks or ink pairs to The resuts show that using a singe Hamitonian path incurs high backup hop cost, due to the ong backup paths invoved. The ADPC scheme reduces this cost significanty by empoying a set of smaer p-cyces. The FRRPC and NEPC schemes use even smaer cyces, reducing this cost further to the eve of pure FRR. We aso observe that the effect of the traffic pattern on the resuts is reativey sma for the FRRPC and NEPC that use short cyces. However, the backup hop cost more directy depends on the traffic pattern for the Hamitonian and ADPC schemes that empoy onger cyces, especiay for scenarios 2 and 3. The ess connected network ike Havana aso has the higher hop cost than Cost-239 and is even higher than USA in scenarios of 1 and 3, because it needs to put more bandwidth for each span, and the backup paths are often more than 3 hops Labe entry overhead Tabe 2 compares the protection schemes in terms of the number of additiona abes needed under the three protection scenarios we consider. For pure FRR, each ink or node is protected independenty of others by estabishing a separate protection LSP per traffic component. Hence, the number of abes is proportiona to the number of traffic components and the tota ength of a backup paths

9 478 C. Cao et a. / Int. J. Eectron. Commun. (AEÜ) 67 (2013) Tabe 2 Labe entry overhead comparison. Topoogy FRR Hybrid FRR/p-cyce Hamitonian ADPC NEPC FRRPC Scenario 1: ink protection Cost Havana Nu 198 USA Scenario 2: node protection Cost Havana Nu 178 USA Scenario 3: ink and node protection Cost Havana Nu 386 USA in the network. As a resut, the number of protection abe entries for the USA topoogy is much higher than in Hanava topoogy, and Havana is aso higher than Cost-239 topoogy. In addition, in order to protect both ink and node, pure FRR needs to depoy two different mechanisms. Therefore, the tota number of abe entries for scenario 3 is the sum of the abe entries required under scenarios 1 and 2. For the hybrid FRR/p-cyce schemes, a affected traffic is forwarded aong on-cyce backup tunnes buit for each p-cyce that can be reaized with ony a few abes. Furthermore, under scenario 3, the Hamitonian, ADPC and NEPC schemes may (re-)use protection abes for both ink and node protection. As a resut, the abe overhead is significanty ower in the hybrid scheme. This is further demonstration of the fact that, by taking a goba design approach in protecting the network inks or nodes, the p-cyce scheme is more efficient in its use of network resources. 6. Concusions We have proposed severa hybrid FRR/p-cyce schemes to impement the oca repair method defined for MPLS networks. These schemes a use backup paths aong a set of pre-configured p-cyces that may be seected using design methodoogies that consider the overa network performance, but otherwise are RFC compiant. Numerica resuts indicate that using a set of reativey short p-cyces outperforms pure FRR in terms of backup capacity and abe overhead, and is comparabe to pure FRR in terms of backup hop cost. These benefits can be reaized for both ink and node protection, and become more significant as the network size grows. Our main concusion is that p-cyce designs are an attractive aternative for MPLS network operators. References [1] Jorge L, Gomes T. The coming of age of MPLS. IEEE Commun Mag 2011;49(Apri (4)): [2] Ashaer H, Emirghani J. Mutiayer dynamic traffic grooming with constrained differentiated resiience in IP/MPLS-over-WDM networks. IEEE Trans Netw Serv Manage 2012;9(March (1)): [3] Cao C, Rouskas G. Hybrid FRR/p-cyce MPLS ink protection design. In: Proceedings of GLOBECOM December p [4] Bigos W, Cousin B, Gossein S, Le Fo M, nakajima H. Survivabe MPLS over optica transport networks: cost and resource usage anaysis. IEEE Seect Areas Commun 2007;25(June (6)): [5] Ruiz M, Pedroa O, Veasco L, Caregio D, Fernadez-Paacios J, Junyent G. Survivabe IP/MPLS-over-WSON mutiayer network optimization. IEEE/OSA J Opt Commun Netw 2011;3(August (8)): [6] Liu M, Tornatore M, Mukherjee B. New strategies for connection protection in mixed-ine-rate optica WDM networks. IEEE/OSA J Opt Commun Netw 2012;3(March (9)): [7] Stamateakis D, Grover W. IP ayer restoration and network panning based on virtua protection cyces. IEEE Seect Areas Commun 2000;18(October (10)): [8] Pan P, Swaow G, Atas A. Fast reroute extensions to RSVP-TE for LSP tunnes. RFC 4090; May [9] Yadav R, Yadav R, Singh H. Enhanced intercyce switching in p-cyce survivabiity for WDM networks. IEEE/OSA J Opt Commun Netw 2010;2(November (11)): [10] Kiaei M, Assi C, Jaumard B. A survey of the p-cyce protection method). IEEE Commun Surv Tutor 2009;11(3rd Quarter (3)): [11] Keng J, Reed M. Bandwidth protection in MPLS networks using p-cyce structure. In: Proceedings of DRCN October. p [12] Keng J, Reed M. Network protection for mesh networks: network coding-based protection using p-cyces. IEEE/ACM Trans Netw 2010;18(February (1)): [13] Schupke DA. Automatic protection switching for p-cyces in WDM networks. Opt Switch Netw (OSN) 2005;2(May (1)): [14] Gangxiang S, Grover WD. Extending the p-cyce concept to path segment protection for span and node faiure recovery. IEEE Seect Areas Commun 2003;21(October (8)): [15] Kodian A, Grover WD. Faiure-independent path-protecting p-cyces: efficient and simpe fuy preconnected optica-path protection. J Lightwave Techno 2005;23(Octorber (10)): [16] Onguetou D, Grover WD. A new insight and approach to node faiure protection with ordinary p-cyces. In: Proceedings of IEEE internationa conference on communications p [17] Jaumard B, Li H. Minimum CAPEX design of segment p-cyces with fu node protection. In: Proceedings of th internationa conference on Optica Network Design and Modeing (ONDM). Apri p [18] Jaumard B, Li H. Segment p-cyce design with fu node protection in WDM mesh networks. In: Proceedings of th IEEE workshop on Loca Metropoitan Area Networks. October p [19] Li L, Buddhikot M, Chekur C. Routing bandwidth guaranteed paths with oca restoration in abe switched networks. IEEE Seect Areas Commun 2005;23(February (2)): [20] Farre A, Ayyangar A, Vasseur JP. Inter-domain MPLS and GMPLS traffic engineering resource reservation protoco-traffic engineering (RSVP-TE) extensions. RFC 5151; February 2008.