Protection Switching Methods for Point-to-Multipoint Connections in Packet Transport Networks

Transcription

1 Protecton Swtchng Methods for Pont-to-Multpont Connectons n Packet Transport Networks Dae-Ub Km, Jeong-dong yoo, Jong Hyun Lee, Byung Chul Km, and Jae Yong Lee In ths paper, we dscuss the ssues of provdng protecton for pont-to-multpont connectons n both Ethernet and MPLS-TP-based packet transport networks. We ntroduce two types of per-leaf protecton lnear and rng. Nether of the two types requres that modfcatons to exstng standards be made. Ther performances can be mproved by a collectve sgnal fal mechansm proposed n ths paper. In addton, two schemes tree protecton and hybrd protecton are newly proposed to reduce the servce recovery tme when a sngle falure leads to multple sgnal fal events, whch n turn places a sgnfcant amount of processng burden upon a root node. The behavor of the tree protecton protocol s desgned wth mnmal modfcatons to exstng standards. The hybrd protecton scheme s devsed to maxmze the benefts of per-leaf protecton and tree protecton. To observe how well each scheme acheves an effcent traffc recovery, we evaluate ther performances usng a test bed as well as computer smulaton based on the formulae found n ths paper. Keywords: Protecton swtchng, packet transport network, pont-to-multpont, P2MP. Manuscrpt receved Feb. 6, 215; revsed Oct. 7, 215; accepted Oct. 21, 215. Ths work was supported by the ICT &D program of MSIP/IITP (No. B , Terabt optcal-crcut-packet converged swtchng system technology development for the next-generaton optcal transport network). Dae-Ub Km (artkdu@etr.re.kr) s wth the Communcatons & Internet esearch Laboratory, ETI, and wth the Department of Informaton and Communcatons Engneerng, Chungnam Natonal Unversty, Daejeon, ep. of Korea. Jeong-dong yoo (correspondng author, ryoo@etr.re.kr) s wth the Communcatons & Internet esearch Laboratory, ETI, Daejeon, and wth the Department of Engneerng, Korea Unversty of Scence and Technology, Daejeon, ep. of Korea. Jong Hyun Lee (jlee@etr.re.kr) s wth the Communcatons & Internet esearch Laboratory, ETI, Daejeon, ep. of Korea. Byung Chul Km (byckm@cnu.ac.kr) and Jae Yong Lee (jyl@cnu.ac.kr) are wth the Department of Informaton and Communcatons Engneerng, Chungnam Natonal Unversty, Daejeon, ep. of Korea. I. Introducton Packet transport network (PTN) technologes, such as Transport Ethernet and Multprotocol Label Swtchng Transport Profle (MPLS-TP), are rapdly ganng n mportance as they become man solutons n the area of transport networks, whch has tradtonally been based on synchronous dgtal herarchy (SDH) and optcal transport networks (OTNs). The boundares between packet and crcut networks are dsappearng as many tradtonal crcut-swtched applcatons such as voce and vdeo are now beng carred over packet-swtched MPLS or Ethernet networks. ecently, many efforts have progressed to develop operatons, admnstraton, and mantenance and protecton swtchng, whch consttute key technologes for promotng Ethernet and MPLS nto transport networks. Such actvtes have manly been conducted n the Internatonal Telecommuncaton Unon- Telecommuncaton Standardzaton Sector (ITU-T) Study Group 15 n close collaboraton wth the Insttute of Electrcal and Electroncs Engneers and the Internet Engneerng Task Force (IETF). Lnear protecton swtchng provdes actve measures to react aganst cable cuts, node falures, or sgnal degradaton on an end-to-end connecton, whch can be seen as an Ethernet Vrtual Local Area Network (VLAN) or MPLS label-swtched path (LSP). For Ethernet lnear protecton, the Automatc Protecton Swtchng (APS) protocol s specfed n the ITU-T ecommendaton G.831 [1], whereas the Automatc Protecton Coordnaton (APC) protocol s specfed n both the IETF FC 7271 [2] and the ITU-T ecommendaton G.8131 [3] for MPLS-TP. The APC protocol was created to enhance the Protecton State Coordnaton (PSC) protocol specfed n the IETF FC 6378 [4] and to provde operator control and 18 Dae-Ub Km et al. 216 ETI Journal, Volume 38, Number 1, February 216

2 experence that more closely models the behavor of lnear protecton seen n the APS protocol. A detaled descrpton of the APC protocol and ts relatonshp wth the APS and PSC protocols can be found n [5]. Ethernet rng protecton (EP), whch has been developed as part of the ITU-T ecommendaton G.832 [6], can support both pont-to-pont (P2P) and pont-to-multpont (P2MP) servces, but t s optmzed for a rng topology utlzng generc mechansms nherted from the tradtonal Ethernet frame header and brdge functons [7]. It s also worth mentonng that EP s well known for havng ssues wth the Flterng Database (FDB) flush operaton, whch causes all rng nodes to broadcast data frames untl source address learnng s complete. To obtan stable protecton swtchng performance, some schemes solvng network stablty ssues related to FDB flush and EP have been dscussed n [8] [12]. Wth the popularty of vdeo dstrbuton, IPTV, and other one-to-many applcatons, both servce provders and network operators need ther premum P2MP servces to be protected. A wde varety of P2MP servces can be effcently realzed over P2MP connectons n Ethernet and MPLS-TP-based PTN. To provde reslency for P2MP connecton networks, protecton swtchng capablty, as seen n other topologes, should be provded to recover traffc wthn the typcal transport network protecton tme requrement (that s, below 5 ms) n the event of a network defect. Ths paper focuses on protecton swtchng methods for a P2MP connecton that s transported over an Ethernet or MPLS- TP network. The protecton swtchng schemes are desgned to provde protecton over a multpont servce, whch s commonly referred to as an Ethernet-Tree servce [13], connectng one root and a set of leaves, but preventng the leaves from communcatng drectly wthout passng through the root. In ths paper, we propose four schemes to provde protecton swtchng capabltes for P2MP connecton networks per-leaf protecton wth exstng lnear protecton, per-leaf protecton wth exstng rng protecton, tree protecton, and a hybrd of per-leaf and tree protectons. Whle two per-leaf protectons can be used wthout any modfcaton to the exstng protecton standards, ther performances can be mproved by a collectve sgnal fal (C-SF) mechansm one that s to be proposed n ths paper. Tree protecton s proposed to reduce the servce recovery tme when a sngle falure leads to multple sgnal fal (SF) events and results n a root node havng to bear a sgnfcant processng burden. A hybrd of per-leaf and tree protectons s also newly proposed to maxmze the benefts of per-leaf and tree protectons. To observe how well each scheme acheves effcent protecton n a falure event, we evaluate ther performance usng a test bed wth real systems and computer smulatons based on mathematcal formulae. The remander of ths paper s dvded nto three sectons. Secton II presents a reference network model and protecton swtchng schemes for reslent P2MP networks. Performances of the presented schemes are evaluated n Secton III, and fnally, conclusons are drawn n Secton IV. II. eference Network Model and Protecton Swtchng Schemes for eslent P2MP Networks A reference network for a reslent P2MP servce network can be modelled as shown n Fg. 1. A root node,, and a set of leaf nodes (,, L9) are connected over a logcal tree structure, whch may branch out at ntermedate nodes (I1,, I4). A protecton tree (dashed lnes) provdes protecton aganst a falure n a workng tree (sold lnes), whch s used for traffc delvery n a fault-free stuaton. The two trees are completely dsjonted to prevent a sngle pont of falure and preconstructed pror to a falure for fast traffc recovery. In the followng subsectons, we propose four protecton swtchng schemes for P2MP networks. The man desgn objectve of the proposals s to reuse any exstng protecton swtchng technologes as much as possble. 1. Per-leaf Protecton wth Exstng Lnear Protecton A P2MP network can be protected wth multple P2P lnear protectons. As shown n Fg. 2, one P2P workng path and one P2P protecton path are prepared for each leaf node. For example, the workng path between the root node and leaf node, -I1-I3-, s backed by the protecton path -I2-I4-. The number of lnear protecton processes (LPPs) pertanng to the root node s equal to the number of leaf nodes, whereas that for a leaf node s equal to only one. A par of LPPs one belongng to a leaf node and one to the root node provde P2P lnear protecton to the correspondng workng and I3 I1 I4 Backbone network Fg. 1. eslent P2MP servce network. I2 P2MP network L3 L4 L5 L6 L7 L8 L9 ETI Journal, Volume 38, Number 1, February 216 Dae-Ub Km et al. 19

3 oot node oot node oot node I1 I2 I1 I2 I1 I2 I1 I2 I1 I2 I1 I2 I1 I2 I1 I2 I1 I2 I3 I4 I3 I4 I3 I4 I3 I4 I3 I4 L3 L4 L5 L6 L7 L8 L9 : LPP : Workng path : Protecton path Fg. 2. Per-leaf protecton scheme. : MEP : LPP : Brdge and selector : SF notfcaton (a) : Set of LPPs : SF collector : C-SF fal notfcaton (b) Fg. 3. Internal operatons of per-leaf protecton: (a) natve mechansm and (b) C-SF mechansm. protecton paths and may be chosen ndependently of other exstng LPP pars. When a defect occurs at any lnk or node along a workng path, the correspondng protecton path s used to delver traffc. We call ths scheme per-leaf protecton n ths paper. In per-leaf protecton, the exstng Ethernet and MPLS-TP lnear protectons can be used wthout the need for modfcaton to Ethernet and MPLS-TP envronments, respectvely. Some operaton examples of Ethernet lnear protecton and MPLS-TP lnear protecton are llustrated n the appendces of ITU-T ecommendaton G.831 [1] and IETF FC 7271 [2], respectvely. Consderng the fact that both Ethernet and MPLS-TP lnear protecton technologes can provde a sub 5 ms protecton swtchng tme regardless of the number of ntermedate nodes between two end nodes as long as ther dstance s less than 1,2 km, per-leaf protecton can also provde carrer-grade reslency for P2MP networks wthout any modfcaton to the exstng standardzed mechansms. However, dependng on the locaton of falure n a network, per-leaf protecton can suffer a performance ssue. When a cable-cut or a node falure occurs near leaf nodes and the number of affected paths s rather small, all LPPs can complete ther protecton swtchng operaton wthn the requred protecton swtchng tme. On the other hand, f the problem occurs near the root node, then any traffc recovery may be delayed due to the sgnalng assocated wth excessve smultaneous SF trggers. Fgure 3(a) shows detaled operatons nsde the root and leaf nodes when a problem occurs at the lnk between the root node and ntermedate node I1. For the sake of brevty, ntermedate nodes are omtted from the fgure. To montor the contnuty of a path between the root node and a leaf node, a par of Mantenance Entty Group End Ponts (MEPs) s actvated at the end of each path [14]. If an MEP detects, va a contnuty check (CC) functon, an anomaly that results n a loss of contnuty (LOC) defect, then t nforms an LPP of an SF condton that has been detected. Then, the LPP runs ts protecton swtchng algorthm to determne the poston of the selector and brdge to swtch over traffc, and generates APS or APC protocol messages to coordnate the swtchng acton wth the other end. The SF notfcaton event s normally executed by an nterrupt-drven algorthm and has hgher prorty than general perodc tasks n mplementatons. If a defect occurs along the -I1 lnk n Fg. 1 that smultaneously affects all of the paths that consttute the workng tree, then all of the LPPs need to be notfed of such an SF. The subsequent mass of sequental operatons that follows ncreases node processng burdens and delays traffc recovery n mplementatons. In partcular, multple nter-processor communcatons (IPCs) between MEPs and LPPs n dfferent lne-card processors negatvely affect the recovery performance. To releve the burden related to IPC for SF notfcatons, all SFs occurrng wthn a certan tme nterval are collected nto a sngle notfcaton and delvered to the place where all of the LPPs resde. As shown n Fg. 3(b), an SF collector performs the C-SF mechansm. After the C-SF notfcaton arrves at the set of LPPs, each of the collectve SFs wthn the sngle notfcaton s processed ndvdually. In Secton III, the performance benefts of the C-SF mechansm are evaluated. 2. Per-leaf Protecton wth Exstng ng Protecton Another possble way to protect a P2MP connecton s to set up multple P2P connectons wth multple rng protecton algorthms, as shown n Fg. 4. In Fg. 4, we assume that nne EP nstances are used for nne rngs; the frst fve rngs consst of sx nodes (-I1-I3-L-I4-I2) and the other four rngs consst of four nodes (-I1-L-I2), where L s leaf node on each rng. Any one of the nodes on a rng can be confgured as the ng Protecton Lnk (PL) owner node. In Fg. 4, the root node s assgned to be the PL owner and the I2 node s confgured as 2 Dae-Ub Km et al. ETI Journal, Volume 38, Number 1, February 216

4 oot node oot node I1 I2 I1 I2 I3 I4 I3 I4 L3 L4 L5 L6 L7 L8 L9 : PL block : PP Fg. 4. EP applcaton for per-leaf protecton. the PL neghbor node. Then, the lnk -I2 s logcally blocked to prevent a loop on the rng under normal operatng condtons. When a rng node detects an SF condton, whch occurs at the lnk drectly connected to ts ports, the rng node blocks the traffc on the faled rng port and transmts ng-aps messages to ndcate the presence of the SF condton. Upon recevng the SF messages, the PL owner unblocks the PL block. The PL neghbor also unblocks the PL block f t s confgured to place the PL block n a normal operatng condton. Detaled operatons of EP can be found n [7]. In a manner smlar to that of the lnear protecton applcaton for per-leaf protecton, ths method requres that multcast servce packets be replcated and sent to multple P2P connectons at the root node; moreover, the method has very poor scalng characterstcs and consumes vast amounts of bandwdth resources. 3. Tree Protecton In tree protecton, a dedcated-protecton rooted multpont connecton (protecton tree) s prepared to back up a workng rooted multpont connecton (workng tree). Any falure on a workng tree, even f the falure affects only a porton of the leaf nodes, wll cause all traffc flowng on the workng tree to be swtched to the protecton tree. As shown n Fg. 5, tree protecton requres a sngle tree protecton process (TPP) n the root node and one TPP n each leaf node. The brdge and selector n a root node and those n each leaf node are coordnated by a protecton protocol to hold the same brdge and selector postons. Protecton protocol messages should be communcated between a root node and each leaf node to coordnate ther brdge and selector postons. As n all the exstng protecton protocols, the protecton protocol messages are necessary to delver the network operator s commands and the SF detected by an end; that s, L3 L4 L5 L6 L7 L8 L9 : Workng tree : Protecton tree Fg. 5. Tree protecton scheme. undrectonal falure. A multcast protecton protocol message generated by the root node s delvered to all the leaf nodes, whle the protocol messages generated by each leaf node are delvered only to the root node. The protecton protocol messages from each leaf node should be dentfable to the root node. Ths can be done by assgnng dfferent connecton IDs, such as VLAN IDs for Ethernet and LSP IDs for MPLS-TP, or by ncludng a node ID from whch the root node can dentfy the source of the protecton protocol message. If a root node ntates protecton swtchng, then tree protecton wll be completed by a sngle message exchange between the root node and all the leaf nodes, and the exstng sngle-phase lnear protecton protocols can be used as s. However, f the protecton swtchng s ntated by a leaf node, then t requres two message exchanges to complete tree protecton swtchng. In the followng subsectons, we show how the exstng Ethernet lnear protecton protocol can be modfed to support tree protecton n cases of SFs on the workng tree. Although we restrct the scenaros n ths paper due to the page lmt, more complex scenaros and a complete state machne can be devsed based on the followng scenaros. In MPLS-TP envronments, the APC protocol can be modfed smlarly. A. Case of Detectng SFs at Leaf Nodes Fgure 6(a) shows an example scenaro to explan the tree protecton operaton ntated by a leaf node. Intally, the network s n a normal condton and traffc flows on the workng tree. When a leaf node, for example, detects an SF on the workng path, all the root and leaf nodes wll swtch to the protecton tree. The operatonal sequence for ths example scenaro s llustrated n Fg. 6(b) and summarzed as follows: 1) All the root and leaf nodes are operatng under normal condtons and usng the workng tree (w) for traffc delvery, TPP ETI Journal, Volume 38, Number 1, February 216 Dae-Ub Km et al. 21

5 whch s ndcated by the no request messages n the fgure. 2) detects an SF on workng. 3) swtches to the protecton tree (p) by settng the brdge and selector (b/s) to the protecton tree, and sends an SF(p) message to the root node. 4) The root node swtches to the protecton tree and sends a new message, X(p), to all the leaf nodes (,, ). 5) Upon recevng message X(p) from the root node, takes no acton and keeps sendng SF(p), but all the other leaf nodes,,,, swtch to the protecton tree and send confrmaton message to the root node. The purpose of the new message X(p) s to request protecton swtchng to all the leaf nodes bar the one that actually ntated the protecton swtchng. For the leaf node who ntated the protecton swtchng ( n ths example), the message X(p) can be consdered as a confrmaton. In a tree protecton archtecture, the protecton swtchng acton s determned by the request havng the hghest prorty wthn the protected doman. For ths, the root node should propagate the Set b/s to p; send confrm & request Workng tree Protecton tree (a) Fg. 6. Tree protecton operaton ntated by leaf node: (a) archtecture, (b) sequence of protecton swtchng operaton, and (c) sequence of reverson operaton. Workng tree Protecton tree SF on w X(p) SF(p) Set b/s to p; send request X(p) X(p) Set b/s to p; send confrm (b) X SF(p) SF(p) Workng SF(p) recovered Start WT Send tmer no request WT(p) WT(p) WT tmer expres Set b/s to w; send request Set b/s to w; Set b/s to w; send confrm send confrm (c) SF(p) Set b/s to p; send confrm WT(p) Set b/s to w; send confrm hghest prorty request n the protected doman by comparng the prorty of remote requests from each leaf node wth ts local request (top prorty local request), and sendng the hghest one (top prorty global request) to each leaf node. In the above example, the message X(p) should be SF(p), whch s devated from the exstng APS protocol. In the APS protocol, the root node would send as t has no defect detected locally and sends and receves traffc to and from the protecton tree. Accordng to the current operatonal prncple of APS Ethernet lnear protecton protocol, a node sends the top prorty global request only when t s the top prorty local request, and sends when the top prorty global request s a remote request from the far-end node. A root node always sends ts top prorty global request regardless of ts orgn. It should be noted that ths change suggested n ths paper can cause leaf nodes to recognze the propagated request from the root node as the local request of the root node. In the above example, all the leaf nodes wll consder the receved SF(p) as f the root node detected an SF n the drecton of any leaf node to the root node. In b-drectonal protecton swtchng, ths forged SF message at a leaf node does not affect the Ethernet lnear protecton APS protocol operaton n the leaf node except for the recovery case. Fgure 6(c) s a contnuaton of Fg. 6(b) and shows the operatonal sequence for the reverson case when the workng tree s recovered from the falure. The operatonal sequence of Fg. 6(c) s summarzed as follows: 1) All the root and leaf nodes are n protecton mode due to an SF on the workng tree at (root node s propagatng SF(p) to all the leaf nodes). 2) detects clearance of SF on workng tree. 3) sends to the root node. 4) Upon recevng, the root node starts a wat-torestore (WT) tmer, whch s used to avod chatterng of selectors n the case of ntermttent defects, and sends WT(p) to all the leaf nodes. 5) When the WT tmer expres, the root node swtches to the workng tree and sends to all the leaf nodes. 6) Upon recevng, each leaf node swtches to the workng tree and sends a confrmaton to the root node. It should be noted that the root node runs a WT tmer when ts top prorty global request s changed from an SF to a no request despte the fact that the SF s not ts local request. A leaf node does not start ts WT tmer as t does not receve from the root node after the recovery B. Case of Detectng SFs at oot Node If a root node ntates protecton swtchng, then tree 22 Dae-Ub Km et al. ETI Journal, Volume 38, Number 1, February 216

6 protecton wll be completed by a sngle message exchange between the root node and all the leaf nodes. In ths case, the APS Ethernet lnear protecton protocol can be used wthout any modfcatons. Fgure 7(a) shows an example scenaro to explan the tree protecton operaton ntated by a root node. Intally, the protected doman s n a normal condton and traffc s flowng on the workng tree (Fg. 7(a)). When a root node detects an SF on the workng path from a leaf node (for example ) to the root node, all the root and leaf nodes wll swtch to the protecton tree (Fg. 7(b)). The operatonal sequence for ths example scenaro s llustrated n Fg. 7(b) and summarzed as follows: 1) Intally, all the root and leaf nodes are n normal condton, whch s ndcated by the presence of messages. 2) The root node detects an SF on the workng tree. 3) The root node swtches to the protecton tree and sends a request SF(p) to all the leaf nodes. 4) Each leaf node swtches to the protecton tree and sends to the root node. When the workng tree recovers from the falure, the protected doman wll revert to the workng tree after WT tmer expraton f the doman s confgured wth a revertve mode, or keep the protecton tree f the doman s confgured wth a nonrevertve mode. As a contnuaton of Fg. 7(b), Fg. 7(c) shows the operatonal sequences for the case of revertve mode. The operatonal sequence of Fg. 7(c) s summarzed as follows: 1) All the root and leaf nodes are n protecton mode due to an SF on the workng tree at the root node. 2) The root node detects clearance of the SF on the workng tree. 3) The root node starts ts WT tmer and sends WT(p) to all the leaf nodes. 4) When the WT tmer expres, the root node swtches to the workng tree and sends to all the leaf nodes. 5) Upon recevng an, each leaf node swtches to the workng path and sends a confrmaton to the root node. 4. Hybrd Scheme of Per-leaf and Tree Protectons Ths protecton mechansm allows flexble operatonal changes between per-leaf and tree protecton schemes. When a falure happens near leaf nodes or some quantty of multple falures occur, a per-leaf protecton mechansm s used. If a defect occurs on a lnk near the root node or many connectons on the workng tree are affected at the same tme, then t may not be able to acheve protecton swtchng wthn 5 ms due to the sgnfcant amount of APS processng burden. At a tme when the number of leaf nodes beng affected by a defect or the number of protecton processes that are runnng smultaneously SF on W Set b/s to p; send request Workng tree Protecton tree (a) Workng tree SF on X Protecton tree Set b/s to p; send confrm SF(p) SF(p) SF(p) Set b/s to p; send confrm (b) SF(p) SF(p) SF(p) Workng recovered Start WT tmer WT(p) WT(p) WT(p) WT tmer expres Set b/s to w; send Set b/s to w; Set b/s to w; request send confrm send confrm (c) Fg. 7. Tree protecton operaton ntated by root node: (a) archtecture, (b) sequence of protecton swtchng operaton, and (c) sequence of reverson operaton. Set b/s to p; send confrm Set b/s to w; send confrm exceeds a certan threshold wthn a certan perod of tme, perleaf protecton s suspended and tree protecton s ntated to reduce APS processng burden. However, when a pror falure exsts n one tree, and the second falure occurs n the other tree, per-leaf protecton s actvated to enhance the network avalablty by utlzng resources on both trees. III. Performance Evaluaton In ths secton, we ntroduce a real test bed n whch the proposed schemes are mplemented and measure ther performances. Then, based on our experence wth the real system, we formulate the restoraton tmes for varous schemes and compare ther performances. An OPNET smulator [15] s used to evaluate the performance of tree protecton. 1. estoraton Tme of Per-leaf Protecton eslence s a key attrbute of PTNs, and the transfer tme must satsfy a sub 5 ms SONET/SDH-grade reslency [16]. ETI Journal, Volume 38, Number 1, February 216 Dae-Ub Km et al. 23

7 Protected traffc restoraton tme (T P ) s the tme from the occurrence of the network mparment to the restoraton of protected traffc and can be expressed as T P = T C + T T + T, (1) where T C (confrmaton tme) s the tme from the occurrence of the network mparment to the nstant when the trggered SF s confrmed as requrng protecton swtchng operatons. It s the sum of the detecton and hold-off tmes. The transfer tme, T T, s the tme nterval between the confrmaton of the SF and the completon of the protecton swtchng operatons, whch nclude settng up the postons of brdge and selector and transmttng any resultng protecton protocol message. The recovery tme, T, s the tme nterval between the completon of protecton swtchng operatons and the full restoraton of protected traffc; ths ncludes the verfcaton of swtchng operatons [17]. Although the sub 5 ms protecton swtchng tme requrement s normally appled to T T n standards [1], [6], our analyss and experments focus on T P to evaluate the performances of varous schemes from the aspect of servce traffc recovery as a whole. To demonstrate the survvablty performances of per-leaf protecton usng Ethernet lnear protecton wth or wthout C-SF, a real test-bed network s bult as n Fg. 8. The expermental network conssts of one root node and two aggregaton leaf nodes that run our proposed schemes, and two ntermedate nodes of commercal Ethernet swtches. The root and leaf nodes are developed based on Broadcom Petra-B, FE6, Altera Arra II GX, and MPC8543V CPU wth Lnux One 1GBASE-L Ethernet lnk s used between a root node and an ntermedate node and sxty 1BASE-SX lnks are placed between each aggregaton leaf node and one of two ntermedate nodes. Each leaf node has 1 lne-cards fve of whch belong to the workng tree and the rest to the protecton tree. Each lne-card s connected to ntermedate nodes wth 12 physcal lnks. The root node s confgured to have up to 1,2 LPPs. Each aggregaton leaf node can handle up to 6 LPPs. The aggregaton leaf node n ths expermental network serves 6 clents, each of whch acts as the leaf node n Fg. 1. For user traffc for each LPP, the traffc generator generates Ethernet frames at an average rate of 1 kfps wth 64 bts per frame. All the user traffc s assumed to be symmetrc and corouted so that the same amounts of traffc between source and destnaton are delvered through the same set of lnks and nodes n each drecton. The length of each lnk s less than 1 m. In our system, a processor s assgned to one task at any tme and the task s executed for one tme slce, whose length s a gven number of the jobs that the task can process at one tme slce. All the LPPs n a node are handled by one task, whch s called LPP task. The length of one tme slce vares n our experments. We consder two knds of queues: the nput queue for the LPP task where the receved SF notfcatons are placed n and the output queue for the MEP task that detects LOC events and sends SF notfcatons. In our mult-slot system, two tasks resde n separate lne cards. The number of SF notfcatons processed by the LPP task n one tme slce s called Q n, whereas the number of SF notfcatons that the MEP task sends va IPC n one tme slot s called Q out. The tme nterval between the completon of one tme slce and the start of the next tme slce for the LPP task and the MEP task, T Qn and T Qout, respectvely, are measured to be 1 ms, on average. In our mplementaton, as the recepton of remote SF messages happens at the same lne-card as the LPP task resdes, no IPC s assumed for the SF messages sent from the remote end node. As shown n Fg. 9, the restoraton tme ncreases wth the number of LPPs n a root node. It also demonstrates that the ncrease of Q n mproves the performance of protecton 1 8 Q n = 3 Q n = 6 Q n = 9 oot node Intermedate nodes estoraton tme (ms) ms restoraton tme requrement Aggregaton leaf node 1 Aggregaton leaf node 2 1 Ggabt Ethernet 6 Ggabt Ethernet Traffc generator & performance analyzer Fg. 8. Expermental setup Number of LPPs Fg. 9. estoraton tme comparson for three duratons of tme slce for LPP task. 24 Dae-Ub Km et al. ETI Journal, Volume 38, Number 1, February 216

8 estoraton tme (ms) Wthout C-SF Wth C-SF, T c-sf =.5 ms Wth C-SF, T c-sf = 3.3 ms Number of LPPs Fg. 1. estoraton tme for C-SF mechansm. swtchng. The number of LPPs that meet the 5 ms restoraton tme requrement ncreases from 3 to 4 when Q n ncreases from 3 to 9. The value of Q out s fxed to 25 n ths experment. The performances of per-leaf protecton wth and wthout the proposed C-SF mechansm are shown n Fg. 1. In the C-SF mechansm, all the SFs that occur durng a certan tme nterval are collected n one notfcaton and the tme nterval s called T c-sf. When the number of SFs exceeds the maxmum number of SFs conveyed n one notfcaton, M c-sf, a notfcaton s generated wthout watng for T c-sf. For the expermental results shown n Fg. 1, T c-sf s set to ether.5 ms or 3.3 ms, and M c-sf s set to 2. The value of Q n s set to 6. The result demonstrates that the C-SF mechansm reduces the restoraton tme sgnfcantly. The number of LPPs that meet the 5 ms restoraton tme requrement ncreases from 38 to 9 when the proposed C-SF mechansm s used wth.5 ms of T c-sf. 2. Formulaton of estoraton Tme for Per-leaf Protecton wth Lnear Protecton Algorthms Based on our experence wth the system n whch the proposed schemes are mplemented, we formulate the restoraton tme wth smultaneous multple protecton trggers for per-leaf protecton wth lnear protecton algorthms assumng the network mparment occurs at the workng tree. In addton to the notatons ntroduced n prevous sectons, we use the followng notatons: N: Total number of LPPs at the root node. The same number as the number of leaf nodes, each of whch has one LPP. n: Number of the affected LPPs that requre traffc swtchover smultaneously, 1 n N. T P (, n): Protected traffc restoraton tme of the th LPP among n affected LPPs. The value s the same for the peerng leaf node to the th LPP. T D (, n): Tme between the occurrence of network mparment and the detecton of an SF at the th workng MEP among n affected workng MEPs n the root node. When the latency of a CC message [14] from the mparment locaton to the root node s T prop_r and the CC message nterval s CC_Perod, the value of T D (, n) s n between 2.5 CC_Perod + T prop_r and 3.5 CC_Perod + T prop_r. T DL (): Tme between the occurrence of network mparment and the detecton of an SF at the workng MEP n leaf node, whch corresponds to the th workng MEP n the root node. When the latency of a CC message [14] from the mparment locaton to the leaf node s T prop_ and the CC message nterval s CC_Perod, the value of T DL () s n between 2.5 CC_Perod + T prop_ and 3.5 CC_Perod + T prop_. T H (, n): Hold-off tme nterval of the th LPP among n affected LPPs. As the same hold-off tme value s used n both end nodes of a P2P connecton, an LPP of a leaf node has the same hold-off tme nterval as ts peer LPP n the root node. T pc (, n): IPC processng tme of the local SF detected at the th workng MEP among n affected MEPs n the root node. T pcl (): IPC processng tme of the local SF detected at the workng MEP n leaf node, whch corresponds to the th workng MEP n the root node. T T (, n): Transfer tme trggered by ether a local SF or remote SF message at the th LPP of the root node. Both local SFs and remote SF messages are processed n the order they arrve. Ths tme s consumed by the root node LPP task that performs the protecton swtchng operaton for the SF. T TL (): Transfer tme trggered by ether a local SF or remote SF message at the LPP of leaf node, whch corresponds to the th LPP of the root node. Ths tme s consumed by the leaf node LPP task that performs the protecton swtchng operaton for the SF. T 2L (, n): End-to-end delay of a protecton protocol message from the th LPP of the root node to the LPP of leaf node, whch corresponds to the th LPP of the root node. Ths value s related to the transmsson speeds and length of lnks and the packet processng tmes n ntermedate nodes. T (, n): End-to-end delay of a protecton protocol message from the LPP of leaf node, whch corresponds to the th LPP of the root node. Ths value s related to the transmsson speeds and length of lnks and the packet processng tmes n ntermedate nodes. ndcates the floor functon. Then, protected traffc restoraton tme, T P, s expressed as T max T (, n). (2) P Dependng on the types of SFs, T P (, n) can be calculated as n the followng subsectons. P ETI Journal, Volume 38, Number 1, February 216 Dae-Ub Km et al. 25

9 A. Case of Undrectonal SFs Detected at oot Node When multple undrectonal SFs occur n the drecton from the leaf nodes to the root node, T (, n) T (, n) T (, n) T ( x, n) P D H pc x a a b TQout TT ( x, n) TQn Q out x b Q n T (, n) T () T (, n), 2L TL (3) where a s the number of SF notfcatons to be sent by the MEP task when the th SF s detected; b s the number of SF notfcatons watng to be processed by the LPP task at the tme that the th SF notfcaton arrves at the nput queue of the LPP task. B. Case of Undrectonal SFs Detected at Leaf Nodes When multple undrectonal SFs occur n the drecton from the root node to the leaf nodes, T (, n) T () T (, n) T () T () T (, n) P DL H pcl TL c TT x n TQn T2L n x c Q n (, ) (, ), where c s the number of SFs watng to be processed by the LPP task of the root node at the tme that the SF message from leaf node arrves at the nput queue of the LPP task. C. Case of Bdrectonal SFs Detected at Both oot Node and Leaf Nodes When multple bdrectonal SFs occur n both drectons, TP(, n) TH(, n) a max TD(, n) Tpc( x, n) TQout x a Q out c TT ( x, n) TQn T2L (, n), x c Q n (4) TDL () TpcL () TTL () T (, n).(5) 3. estoraton Tme Analyss for Per-leaf Protecton wth C-SF Mechansm In ths secton, we consder the restoraton tme of per-leaf protecton wth the proposed C-SF mechansm. Assumng the SF detected at the th workng MEP among n affected workng MEPs n the root node s collected n the jth collectve SF notfcaton, T P (, n) for per-leaf protecton wth C-SF mechansm when multple bdrectonal SFs occur n both drectons, s expressed as (6). In (6), T Wc-sf (, n) s the tme between the detecton of an SF at the th workng MEP among n affected workng MEPs n the root node and the completon of the jth collectve SF notfcaton ready to be sent out va IPC, T Wc-sf (, n) T c-sf. The number of collectve SF notfcatons to be sent by the MEP task when the jth collectve SF notfcaton s formed s represented by d. TP( n, ) TH( n, ) max TD( n, ) TWc-sf( n, ) d T x n T T ( x, n) j j pc (, ) Qout T xjd Q j out xc c TQn T2L (, n), TDL () TpcL () Q n TTL () T (, n). (6) We omt the expressons of T P (, n) for the cases of undrectonal SFs, as they can easly be derved smlarly. Fgure 11 shows the restoraton tmes calculated from (5) and (6) wth varous values of Q n and T c-sf. We can observe that the graphs match the expermental results shown n Fgs. 9 and 1. We assume that T H (, n) s zero and T D (, n) occurs equally lkely between 2.5 CC_Perod + T prop_r and 3.5 CC_Perod + T prop_r. From our experments, the mean of T pc (, n) shows two dfferent values for = 1 and 2 and the values measure 29 µs and 19 µs for = 1 and 2, respectvely. The mean of T T (, n), measure 87 µs, 32 µs, or 21 µs when (mod Q n ) s 1, 2, or 3, respectvely. 4. estoraton Tme of Tree Protecton and ts Comparson wth other Protecton Protocols For tree protecton, the protected traffc restoraton tme for Sum of confrmaton tme and tranfer tme (ms) Wthout C-SF, Q n = 3 Wthout C-SF, Q n = 6 Wthout C-SF, Q n = 9 Wth C-SF, Q n = 6, T c-sf = 3.3 ms Wth C-SF, Q n = 6, T c-sf =.5 ms Number of LPPs Fg. 11. estoraton tme comparson based on derved equatons. 26 Dae-Ub Km et al. ETI Journal, Volume 38, Number 1, February 216

10 undrectonal SFs detected at the root node can be wrtten as TP TD(1, n) TH Tpc(1, n) (7) T max T (, n) T ( ) T (, n), T 2L TL where T H and T T denote the hold-off tme nterval and the transfer tme at the TPP n the root node, respectvely. The restoraton tme of tree protecton for undrectonal SFs detected at the leaf nodes s wrtten as TP TH mn TDL( ) TpcL( ) TTL( ) T(, n) (8) T max T ( j, n) T ( j) T ( j, n). T 2L TL j We leave formulaton of the restoraton tme of tree protecton for bdrectonal SFs for future work. The performance of protecton swtchng wth a tree protecton protocol and ts benefts are evaluated n comparson wth other protecton swtchng protocols n P2MP connecton networks. To examne transent behavors of dfferent protecton protocols more closely, we rely on an OPNET smulator. As shown n Fg. 12, our computer smulaton network conssts of one root node () wth a server (S), 42 ntermedate nodes, and 1, leaf nodes, each of whch s connected to a clent host. All the lnks connectng the server, the root node, and ntermedate nodes are assumed to have a 1 Gbps capacty and the remanng lnks are assumed to have a 1 Gbps capacty, whch s large enough not to lmt the traffc volume generated by the server and each clent host. Any pars of two adjacent nodes are connected wth 8 km fber lnks, whose propagaton delay s approxmated to 4 µs. Each host generates Ethernet frames at an average rate of 5 kfps destned to the server. The server generates Ethernet frames at an average rate of 5 kfps for each clent host. The lengths of Ethernet frames are exponentally dstrbuted wth a mean of 1 bytes and ther nter-arrval tmes are also exponentally dstrbuted. The CC message rate s always set to 333 fps. Accordng to the standard protocol message transmsson rule [1] [6], APS or -APS message transmsson rates are set to 333 fps n a burst mode, whch s supposed to happen rght after the protecton swtchng occurs, followed by a contnuous mode, where the messages are generated at every 5 s. In the case of hybrd protecton, the threshold value for from per-leaf to tree protecton s set to 35 protecton processes wthn a 6.6 ms nterval. The two graphs n Fg. 13 show the restoraton tmes for all the aforementoned protecton schemes n the cases of bdrectonal SFs and undrectonal SFs n a P2MP connecton network. The results are obtaned by varyng the number of leaf nodes that are affected smultaneously by a sngle network defect. For rng protecton, all the leaf nodes are assumed to be PL owner, and the PL blocks are placed at the lnks n the workng tree. The restoraton tme of rng protecton s longer estoraton tme (ms) Per-leaf ng Tree Hybrd Number of affected leaf nodes (a) 5 ms restoraton tme requrement S 8 Per-leaf ng Tree Hybrd I1 I3 I4 I21 I22 I23 I24 I41 I42 I2 estoraton tme (ms) ms restoraton tme requrement L5 L51 L95 L951 L952 C1 C2 C5 C51 C95 C951 C952 C1 Fg. 12. Smulaton scenaro n normal Ethernet tree topology Number of affected leaf nodes (b) Fg. 13. estoraton tme comparson: (a) bdrectonal SF and (b) undrectonal SF. ETI Journal, Volume 38, Number 1, February 216 Dae-Ub Km et al. 27

11 than that of per-leaf protecton because of the addtonal delay of -APS. Both rng and per-leaf protecton schemes show that the restoraton tmes ncrease as the number of affected leaf nodes ncreases. In contrast, tree and hybrd protecton schemes are stablzed regardless of the number of affected leaf nodes. The performance of the hybrd scheme n terms of restoraton tme cannot be better than that of tree protecton, but t can utlze resources on both trees and enhance the network avalablty more so than tree protecton f the affected leaf nodes are less than the threshold value. When all the 1, leaf nodes are affected by a lnk falure, the data rates measured at the lnk from the root node to the server and the lnks from the leaf nodes to ther clent hosts are shown n Fg. 14. The lnk falure occurs at 1. s, and the data rates are measured every 5 ms. Traffc s recovered n about 2 ms and 3 ms after the falure for the tree and hybrd protecton schemes, respectvely. However, n the cases of rng and per-leaf protectons, ther results exceeded 1 ms. eceved data rate (Gbps) Average receved data rate (Gbps) Per-leaf ng Tree Hybrd Smulaton tme (s) (a) Per-leaf ng Tree Hybrd Smulaton tme (s) (b) Fg. 14. Sngle lnk falure n P2MP topology: (a) data rate at lnk from to S and (b) average data rate at each lnk from leaf node to clent. IV. Concluson We proposed varous protecton swtchng schemes for P2MP connectons n PTN. The man purpose of a P2MP connecton protecton s to acheve fast traffc recovery whle the exstng protecton swtchng technologes can be reused wth mnmal modfcatons. Observng that the IPC tme has a strong nfluence on the restoraton tme, a C-SF mechansm was proposed to releve the burden related to IPC for SF notfcatons n the case of a per-leaf protecton scheme. To maxmze the aglty of protecton swtchng for the P2MP connecton, tree protecton was consdered and ts detaled operatonal behavor was presented. To maxmze the network avalablty subject to the sub 5 ms protecton swtchng tme constrant, a hybrd of per-leaf and tree protecton schemes was also proposed. A comparson study among the presented schemes has been performed, and the hybrd scheme showed well-balanced performance between fault-recovery tme and network avalablty. Although the performance of per-leaf protecton schemes largely depends on the number of protected nstances n a P2MP connecton, the tree and hybrd schemes guarantee fast and relable protecton swtchng regardless of the number of leaf nodes n a P2MP connecton network. eferences [1] ITU-T ec. G.831/Y.1342, Ethernet Lnear Protecton Swtchng, June 211. [2] IETF FC 7271, MPLS Transport Profle (MPLS-TP) Lnear Protecton to Match the Operatonal Expectatons of Synchronous Dgtal Herarchy, Opt. Transport Network, and Ethernet Transport Network Operators, June 214. [3] ITU-T ec. G.8131/Y.1382, Lnear Protecton Swtchng for MPLS Transport Profle (MPLS-TP), July 214. [4] IETF FC 6378, MPLS Transport Profle (MPLS-TP) Lnear Protecton, Oct [5] J.-D. yoo et al., MPLS-TP Lnear Protecton for ITU-T and IETF, IEEE Commun. Mag., vol. 52, no. 12, 214, pp [6] ITU-T ec. G.832/Y.1343, Ethernet ng Protecton Swtchng, Feb [7] J.-D. yoo et al., Ethernet ng Protecton for Carrer Ethernet Networks, IEEE Commun. Mag., vol. 46, no. 9, Sept. 28, pp [8] J.K. hee, J. Im, and J.-D. yoo, Ethernet ng Protecton Usng Flterng Database Flp Scheme for Mnmum Capacty equrement, ETI J., vol. 3, no. 6, Dec. 28, pp [9] K.-K. Lee, J.-D. yoo, and S. Mn, An Ethernet ng Protecton Method to Mnmze Transent Traffc by Selectve FDB Advertsement, ETI J., vol. 31, no. 5, Oct. 29, pp Dae-Ub Km et al. ETI Journal, Volume 38, Number 1, February 216

12 [1] K.-K. Lee, J.-D. yoo, and Y. Km, Impacts of Herarchy n Ethernet ng Networks on Servce eslency, ETI J., vol. 34, no. 2, Apr. 212, pp [11] K.-K. Lee and J.-D. yoo, Flush Optmzaton to Guarantee Less Transent Traffc n Ethernet ng Protecton, ETI J., vol. 32, no. 2, Apr. 21, pp [12] K.-K. Lee, C.-K. Lee, and J.-D. yoo, Enhanced Protecton Schemes to Guarantee Consstent Flterng Database n Ethernet ngs, IEEE Global Commun. Conf., Mam, FL, USA, Dec. 6 1, 21, pp [13] The Metro Ethernet Forum MEF 1.2, Ethernet Servces Attrbutes Phase 2, Oct. 29. [14] J.-D. yoo et al., OAM and ts Performance Montorng Mechansms for Carrer Ethernet Transport Networks, IEEE Commun. Mag., vol. 46, no. 3, Mar. 28, pp [15] OPNET Technologes Inc. Accessed Oct opnet.com [16] M. Huynh, S. Goose, and P. Mohapatra, eslence Technologes n Ethernet, Comput. Netw., vol. 54, no. 1, Jan. 21, pp [17] ITU-T ec. G.88.1, Generc Protecton Swtchng Lnear Tral and Subnetwork Protecton, May 214. focused on next-generaton networks, carrer-class Ethernet, and MPLS-TP technology research, especally partcpatng n OAM and protecton standardzaton actvtes n ITU-T. He s the edtor of G.8131 (MPLS-TP lnear protecton), G.8132 (MPLS-TP rng protecton), and G.88.1 (Generc protecton - Lnear) recommendatons and a vce-charman of ITU-T Study Group 15. He co-authored TCP/IP Essentals: A Lab-Based Approach (Cambrdge Unversty Press, 24). He s a member of the Eta Kappa Nu assocaton. Jong Hyun Lee receved hs BS, MS, and PhD degrees n electroncs engneerng from Sungkyunkwan Unversty, Suwon, ep of Korea, n 1981, 1983, and 1993, respectvely. Snce 1983, he has been wth ETI, where he has served as a drector for both the Optcal Communcaton Department and the esearch Strategy & Plannng Department. He has also served as an executve drector of the Optcal Internet esearch Department. Hs current research nterests are packet-crcut-optcal converged swtchng systems, green Internet data centers, and optcal access networks. Dae-Ub Km s a prncpal researcher at ETI and s currently workng toward a PhD degree n nformaton and communcatons engneerng at Chungnam Natonal Unversty, Daejeon, ep. of Korea. He receved hs MS degree n nformaton and communcatons engneerng from the Korea Advanced Insttute of Scence and Technology, Daejeon, ep. of Korea, n 21 and hs BS degree n electronc engneerng from Yeungnam Unversty, Gyeongsan, ep. of Korea, n Snce he joned ETI n 21, hs work has been focused on next-generaton networks, wreless backhaul networks, carrer-class Ethernet, OTN, and MPLS-TP technology research, especally partcpatng n protecton standardzaton actvtes n ITU-T. Byung Chul Km receved hs BS degree n electronc engneerng from Seoul Natonal Unversty, ep. of Korea and hs MS and PhD degrees n electronc engneerng from the Korea Advanced Insttute of Scence and Technology, Daejeon, ep. of Korea, n 1988, 199, and 1996, respectvely. From 1993 to 1999, he worked as a research engneer at Samsung Electroncs, Suwon, ep. of Korea. Snce 1999, he has been a professor at the Department of Informaton and Communcatons Engneerng, Chungnam Natonal Unversty, Daejeon, ep. of Korea. Hs research nterests nclude computer networks, wreless Internet, sensor networks, and moble communcatons. Jeong-dong yoo s a prncpal researcher at ETI and a UST professor wth the Department of Engneerng, Korea Unversty of Scence and Technology, Daejeon, ep. of Korea. He holds MS and PhD degrees n electrcal engneerng from the Polytechnc Insttute of New York Unversty, USA and a BS degree n electrcal engneerng from Kyungpook Natonal Unversty, Daegu, ep. of Korea. Upon completng hs PhD n the area of telecommuncaton networks and optmzaton, he started workng for Bell Labs, Lucent Technologes, NJ, USA, n Whle he was wth Bell Labs, he was manly nvolved wth performance analyss; evaluaton; and enhancement study for varous wreless and wred network systems. Snce he joned ETI n 24, hs work has been Jae Yong Lee receved hs BS degree n electroncs engneerng from Seoul Natonal Unversty, ep. of Korea and hs MS and PhD degrees n electronc engneerng from the Korea Advanced Insttute of Scence and Technology, Daejeon, ep. of Korea, n 1988, 199, and 1995, respectvely. From 199 to 1995, he worked as a research engneer at the Dgcom Insttute of Informaton and Communcatons, Seoul, ep. of Korea. Snce 1995, he has been a professor at the Department of Informaton and Communcaton Engneerng, Chungnam Natonal Unversty, Daejeon, ep. of Korea. Hs research nterests nclude computer networks, wreless Internet, sensor networks, and moble communcatons. ETI Journal, Volume 38, Number 1, February 216 Dae-Ub Km et al. 29