One-to-Many Multicast Restoration Based on Dynamic Core-Based Selection Algorithm in WDM Mesh Networks

Transcription

1 Proceeings of the International MultiConference of Engineers an Computer Scientists 9 Vol I IMECS 9, March 18 -, 9, Hong Kong One-to-Many Multicast Restoration Base on Dynamic Core-Base Selection Algorithm in WDM Mesh Networks I-Shyan Hwang, *San-Nan Lee, Zen-Der Shyu an Kang-Peng Chen Abstract-This paper proposes a novel ynamic core-base selection (DCS) algorithm for the multicast restoration in WDM mesh networks. The core-base fault tolerance scheme provies a flexible way to control a number of core noes with less control overheas for searching the routing path, wavelength assignment (RWA) an restoration paths when fault occurs in the one-to-many multicast omain. Compare with the source-base scheme, core-base schemes are easier to maintain, an specifically scalable in large-scale topologies. In the core-base fault tolerance scheme, k-tuple omination noes are selecte to form a minimum size vertex subset such that each vertex in the graph is ominate by at least k vertices, where the k is efine as two in this paper. The propose DCS algorithm is efine as each noe in multicast tree session must be irectly connecte to at least one core noe in multicast tree session an also has to be irectly connecte to at least one core noe out of multicast tree session. The primary aim of this work is to provie the scalable an fast local survivability base on the information from core noes. Simulation results show that the propose algorithm outperforms the Dual Tree an MRLR algorithms in terms of total hop counts neee for all recovery paths an blocking probability for ifferent network topologies. Keywors: DCS, Multicast restoration, Local survivability, Dual Tree, MRLR. I. Introuction Optical networks with Dense Wavelength Division Multiplex (DWDM) can provie multiple ata channels to supply high spee, high capacity to perform banwith-intensive multicast transmission service [1] technology, which uses wavelength ivision, markely increases the banwith of existing optical fiber networks. Multicasting is the simultaneous elivery of ata stream from one or more sources to multiple receivers by using one or more tree-base structures in the network. The benefit of multicast is that the source sens a single copy to the entire estinations, such that multicast can utilize banwith efficiency. In general, the one-to-many multicast routing algorithm can be classifie as source-base moe an core-base moe. In source-base moe, multicast tree roote at the source noe an connecte to each member in the multicast session [2]. Then, ata packets originating from I-Shyan Hwang is a professor in the Department of Computer Science & Engineering at the Yuan-Ze University, Chung-Li, Taiwan (ishwang@saturn.yzu.eu.tw). San-Nan Lee is a assistant professor in the Department of Computer Science an Information Engineering at Vanung University, Chung-Li, Taiwan (Tel: ~727, Fax: ~7, gran@vnu.eu.tw). Zen-Der Shyu is currently pursuing the Ph. D egree at the Yuan-Ze University. Since 199, he serve as an instructor in the Army Acaemy R.O.C. Kang-Peng Chen receive M.S. egree in Department of Computer Science an Engineering from the Yuan-Ze University, Chung-Li, Taiwan in 8. the source noe are sent to all the estination noes via the links of multicast tree. The Distance Vector Multicast Routing Protocol (DVMRP) [3] is a ense moe routing protocol which has goo performance in network environments with high banwith an ensely istribute multicast session members. In heterogeneous Internet environment, it potentially has to support many thousans of multicast groups, each of which may be sparsely istribute, so this technique oes not scale well. The Protocol Inepenent Multicast-Dense Moe (PIM-DM) [4] is an efficient protocol when most receivers are intereste in the multicast ata, but oes not scale well across larger omains in which most receivers are not intereste in the ata. The main ifference between DVMRP an PIM-DM is that the PIM-DM introuces the concept of protocol inepenence such that any unerlying unicast routing protocol to perform reverse path forwaring checks. The Multicast Open Shortest Path First (MOSPF) [] is an extension to the Open Shortest Path First (OSPF) protocol to support multicast routing an allows routers to share information about group memberships. In core-base moe, some noes for each group are selecte as the core noes an multicast tree roote at core noe an constructe to span all the group members. Data packets flow from any source to its parent an chilren, such as Core-base Tree (CBT) [6] an Protocol Inepenent Multicast-Sparse Moe (PIM-SM) [7]. The CBT may buil one or multiple core-base biirectional trees which are share by all of the group's seners an receivers with core noe selection algorithm. However, the traffic may be concentrate at core noes, the scalability an more management cost shoul be consiere. The PIM-SM is an efficient routing protocol to multicast groups that span all estinations istribute sparsely in inter-omain network environment. The core-base scheme is preferable to source-base for multiple sources in the multicast group. The avantage of core-base scheme has less control overhea of a single share tree rather than multiple trees. Recent researches in core-base scheme focus on the multi-core selection by k-center or r-ominating in multicast trees [8]. Core noe selection algorithms can be classifie as (1) one core noe, an (2) multiple core noes selection [9]. The rest of this paper is organize as follows. Section II escribes relate work of survivability scheme in multicast. Next, Section III proposes a novel ynamic core-base selection (DCS) algorithm for one-to-many multicast restoration in WDM mesh networks, which is capable of proviing overall protection for optical noes an fibers. Section IV compares the system performance of the propose algorithm with the Dual Tree [] an Multiple Ring-base Local Restoration (MRLR) [] algorithms in terms of total hop counts neee for all ISBN: IMECS 9

2 Proceeings of the International MultiConference of Engineers an Computer Scientists 9 Vol I IMECS 9, March 18 -, 9, Hong Kong recovery paths an blocking probability for ifferent network topologies. Finally, conclusions are rawn in Section V. II. Multicast Fault Tolerance With the intensive banwith in DWDM optical fibers, a network function failure, such as network equipment corrupte or fiber cut, will cause serious ata loss. For protecting multicast session against failure, network survivability must be consiere in esigning DWDM networks. The light-tree protection for multicast has more challenging than the unicast in optical networks, since a link or a noe failure will affect several estinations at the same time. So keeping multicast session functionally nees a heuristic proceure an the objective is to achieve high reliability an fast recoveries with minimum backup resource an cost in the multicasting scheme. The preplanne fault tolerance with backup rerouting algorithm can provie fast recovery time, an it is suitable for one-to-many multicast with ifferent backup structures belonging to the reuce topology which incluing a set of backup paths to minimize the multicast tree cost after recovery. There are three categories in multicast fault tolerance schemes as shown in Fig. 1. Fig. 1 Three categories of multicast fault tolerance algorithms. Ring-base protection: Ring-base protection has been part of the transport network lanscape for some time ue to the wie eployment of ring network topologies. The first mesh-ring ecomposition is known as the noe cover [12], in which the set of rings is selecte such that each noe belongs to one or more rings an each link belongs to no more than one ring. A noe cover oes not necessarily cover all the links an the uncovere links remain unprotecte. This problem can be eliminate using the ring cover []. Each link must belong to at least one ring an all links are protecte, however, spare resources may be reunant. Unfortunately, eciing the minimal ring cover in non-planar networks is an NP-har problem an is not scalable. This type of algorithm can, therefore, recover quickly from failures an etermine how to allocate recovere traffic loas base on current traffic loa an the network banwith along the restoration paths. Tree-base protection: Link protection an path protection [14] are parts of the tree-base protection. The link protection is to fin a backup link between noes, but it causes a huge cost an waste banwith. The path protection establishes a isjoint path between each source-estination pair. The reunant tree protection is propose in [1] to establish a new multicast tree from original source to the estination noes bypassing the existing tree noes. In ual tree protection [], affecte leaf noes are connecte to an unaffecte leaf noe. It requires that the unerlying network topology is a bi-connecte graph, in which there are at least two vertex-isjoint or link-isjoint paths between any two noes. Four multicast tree fault tolerance schemes are propose in [16], namely, Group-base protection: The concept of Share Risk Link Group (SRLG) [17] is a group of network links that share a common physical resource. After an initial failure an before reprovisioning, it cannot support restoration for aitional failures to other light-paths in the same restoration group. Therefore, two or more working paths uner the same failure risk cannot share the same protection resource. The Share Banwith Assignment (SBA) [18] provies the ability using integer linear programming to allocate banwith an fin backup segments by ynamic programming when several multicast working paths sharing the common backup path. The Share Segment Protection with Reprovisioning (SSPR) [19] is similar to the SRLG that the SSPR can provie ual links fault tolerance an increase the banwith utilization. The propose ynamic core-base selection (DCS) algorithm is efine as each noe in multicast tree session must be irectly connecte to at least one core noe in multicast tree session an also has to be irectly connecte to at least one core noe out of multicast tree session. The DCS algorithm is extene to core noes selection which prouces the core-base tree to enhance restoration ability. III. Propose Algorithm In general, fining the k-tuple ominating set (k-ds) of a graph G = { V, E }, where noes set V = { v,, 1 v m } with m noes an links set E = { e1,, en } with n links, that ominates the multicast tree noes in a subset of core noes S, where S V, such that each noe in the subset V \ S is ajacent to at least k core noes in S. The value of k is given by two in this paper. The propose ynamic core-base selection (DCS) algorithm is efine as each noe in multicast tree session must be connecte to at least one core noe in multicast tree session an also the noe has to be connecte to one core noe out of multicast tree session. Each core noe has a core routing table (CRT) which recors the information between current core noe an other core noes, such as hop counts, available wavelengths, an the ominating noes of the current core noe. The available wavelength is use to gather wavelengths usage statistic for each link of recovery path. In aition, each non-core noe in multicast session has a non-core routing table (NCRT) which recors the information of its ominate core noes in terms of available wavelengths. Both tables are broacaste through control channels perioically to other noes for upating the information base on the network status. ISBN: IMECS 9

3 Proceeings of the International MultiConference of Engineers an Computer Scientists 9 Vol I IMECS 9, March 18 -, 9, Hong Kong The propose DCS algorithm is a istribute core selection algorithm which has two subsets: (1) multicast session member ominating set (MDS), (2) non-multicast session member ominating set (NDS). In such a way, each noe in one-to-many multicast session will be ominate at least twice to enhance the restoration ability. Both MDS an NDS are establishe by the DCS algorithm an shown as follows: Dynamic core-base selection (DCS) algorithm Input: Multicast session tree T, network topology Output: MDS an NDS 1. Initialize the MDS={s}, where s is root 2. All parent noes of leaf noes in T are ae to MDS 3. Delete leaf noes an parent noes of leaf noes, an then form new sub tree T 4. If T is not null. Go to Step 2 6. Calculate K={k 1,,k i }, where k i is the number of connections noe i in non-multicast session to the members in multicast session with one hop 7. The noe which has the maximal egree in K is ae to NDS 8. If all multicast session members are ominate by core noes in NDS or no one can be ominate 9. Done. else Go to Step k = k =3 6 k =4 12 k =1 Source = {2} Destinations = {9,12,14} MDS = {2,7, } NDS = {3,,,} 7 8 k 8 =2 Fig. 2 Example of the MDS an NDS The MDS is efine as each noe in the multicast session is ominate at least once by core noes resie in multicast session. Initially, the root noe of multicast session as the source noe is assume to be a core noe, an then fins the core noes from multicast leaf noes to the source noe, recursively. First, the parent noes of leaf noes are chosen as core noes an the leaf noes an parent noes of the chosen core noes are elete to form a new sub tree. This process is to make sure that each noe can be ominate once by core noes. Base on the new sub tree, the parent noes of the new leaf noes are chosen as core noes an the new leaf noes an parent noes of the chosen core noes are elete to form another new sub tree. Repeating the process until the source noe or null noe is left. For example, shown in Fig. 2. The NDS is efine as each noe in multicast session has at least once ominate by core noes from non-multicast session. The selection of core noe from non-multicast session is base on the connectivity for the noes in non-multicast session to the multicast session. Given K = { k1,..., k i }, where k i is the number of connections from noe i in non-multicast session to the members in multicast session with one hop. After calculating the K = { k1,..., k i } of each noe in non-multicast session to the multicast session, the noe with maximum k i is chosen as a core noe. Repeating the process until all multicast session members are ominate or no other core noes can be ominate. For example, shown in Fig. 2, calculating the k i of all noes in non-multicast session to the multicast session, e.g., noe have the k of 4, noe has the k of 3, noe 16 has k 16 of 1 an noe 4 has k 4 of, the other noes have the k, where i = { 1,3,8,,1} of 2. i Then, the noe with k of 4 is chosen as a core noe which can ominate the noes 6, 7, an 12; the noe with k of 3 is chosen as another core noe which can ominate the noes 6, 9 an ; the noe 3 with k 3 of 2 is chosen as a core noe which can ominate the noes 2 an 7, an noe with k of 2 is chosen as a core noe to ominate noes 9 an 14. Then, the NDS inclues noes 3,, an. A restoration scheme for one-to-many multicast communication (RSOMMC) algorithm to resolve the noe or link faults is propose in this session. When fault occurs, the upstream noe u, where u T, ajacent to the faile link or the faile noe will broacast the failure type an location, an etermine which estination noe(s) is affecte base on the CRT from S core or itself if the noe u is a core noe. Each affecte estination noe nees to fin the recovery paths P with minimum hop counts base on the following equation: P = Pv + Pcore + P, where P is the path from upstream noe of the faile v link or the parent noe of the faile noe to the ominating core noe, S core, P core is the path between the ominating core noe, S core, near the fault an core noe ominates the affecte estination noes, D core, an P is the path from the ominating core noe, D core to the affecte estination noes. The P v is null when the upstream noe of the faile link or the parent noe of the faile noe v is a core noe. Noe fault can be seen as multiple links fault an tries to fin the path from the upstream noe of the faile link or the parent noe of the faile noe to each estination noe iniviually. The propose algorithm can maintain multicast session an choose the minimum hop count to be P core. Once the P is etermine, core P v an P are also known then the recovery path is establishe. The pseuo coe of restoration scheme for one-to-many multicast communication (RSOMMC) for link failure or noe failure an the omination fault recovery scheme is escribe as follows: Restoration scheme for one-to-many multicast ISBN: IMECS 9

4 Proceeings of the International MultiConference of Engineers an Computer Scientists 9 Vol I IMECS 9, March 18 -, 9, Hong Kong communication (RSOMMC) Input: Core routing table (CRT), Non-core noe routing table (NCRT) Output: Recovery path 1. If a link fault occurs 2. The upstream noe of the faile link, u, fins the core noe S core that ominate the noe u, an core noe D core_i that ominates the affecte estination noes where i is the sequences of estination noes 3. Accoring the CRT an NCRT, the path P v is selecte from the source noe to S core base on the minimum hop counts, P is the path between S core core an D core_i, an the P is the path between the D core_i an estination noe. 4. If there is another estination noe. Go to step 2 6. else 7. Return P = Pv + Pcore + P 8. If a noe fault occurs 9. Sets all links connecting to the noe are faile. If all links are recovere, then one. else 12. Go to step 2 1. Done Domination fault recovery algorithm Input: MDS, NDS, CRT an NCRT Output: CRT an NCRT 1. If fault occurs 2. Executes the RSOMMC algorithm 3. Execute the ynamic core-base selection (DCS) algorithm to upate the MDS an NDS 4. Each core noe upates the CRT. Each non-core noe upates the NCRT 6. Done Link failure Base on the Fig. 2, a minimum spanning tree with minimum hop counts for multicast session tree is establishe, noe 2 is the source noe an noes 9, 12, an 14 are estination noes. If the link 6- is cut an the affecte estination noes are noes 9 an 14, the source noe of the faile link is noe 6 an the recovery path of source noe to affecte estination noes can be obtaine base on P = Pv + Pcore + P, which is shown Fig k = k =3 6 k =4 12 k =1 Source = {2} Destinations = {9,12,14} MDS = {2,7, } NDS = {3,,,} Fig. 3 Example of link failure Noe failure Base on the Fig. 2, if noe 6 is corrupte, the links 2-6, 1-6, -6, 6-, 6-, 6-7 are isconnecte an the affecte estination noes are 9 an 14. The parent noe of the faile noe is noe 2 an the recovery path of parent noe to affecte estination noes can be obtaine base on P = P + P + P, 7 8 v k 8 =2 core which is shown in Fig k = k =3 6 k =4 12 k =1 Source = {2} Destinations = {9,12,14} MDS = { 2,7, } NDS = {3,,,} Fig. 4 Example of noe failure 7 8 k 8 =2 IV. Performance Evaluation The performance of the propose algorithm is analyze by simulating the USANet (24 noes an 43 links), ChinaNet (39 noes an 72 links) an NTTNet (7 noes an 81 links) networks. In the experiments, each link inclues only one fiber on which 32 an 64 wavelength ata streams can be transmitte, incluing two wavelengths use as biirectional control channels an the conversion time of each converter is 1μs. Source noe an estination noes of multicast session are ranomly selecte, an the multicast group size (session size) is also ranomly assigne from the interval {6, 7,8}. Then, the minimum spanning tree (MST) approach is applie to construct the multicast tree for each multicast session. Multicast connection requests arrive accoring to a Poisson istribution λ = { 1,2,,} per minute, an their holing time is exponentially istribute (approximately a few minutes to hours). After the working tree is constructe for the multicast session, a single failure point (link failure or noe failure) for each multicast session is ranomly set in the multicast tree. Simulation programs are evelope using the OPNET an the performance of the propose algorithm herein is compare with those of the Dual Tree an MRLR algorithms in terms of the average hop count an blocking probability. The average hop count is efine as average hop count of all recovery paths per session an the blocking probability is efine as one or more restoration paths can not be establishe when fault occurs. a. Link failure Average hop count Figure shows the average hop count of link failure versus request rate with the ifferent numbers of channels for three ifferent algorithms in the USANet, ChinaNet, an NTTNet. The propose algorithm has the least average hop count neee than the other two algorithms for ifferent network topologies. Moreover, when the request rate is higher an the channel number is smaller, the average hop count of recovery paths of three algorithms are all increase. It is interesting to notice that the average hop count of recovery paths of the three algorithms are increase when the network topology is more complicate an has more protection paths can be chosen in the USANet. Dual tree protection nees to iscover fully isjoint paths; hence the hop count of each recovery path is larger than other algorithms. It can be also observe that the average hop count of the MRLR is ISBN: IMECS 9

5 Proceeings of the International MultiConference of Engineers an Computer Scientists 9 Vol I IMECS 9, March 18 -, 9, Hong Kong higher than the propose algorithm because the ring-base protection in some situations gets more hop count route even shorter paths exist (a) USANet Fig. 6 Blocking probability versus request rate with Figure 6 compares the blocking probability of link failure versus request rate with ifferent numbers of channels for three ifferent algorithms in the USANet, ChinaNet, an NTTNet. When the request rate is higher an the channel number is smaller, all the blocking probabilities of the three algorithms are increase. However, the blocking probabilities in all algorithms are improve when the network topology becomes complicate an more recovery paths can be chosen. Our propose algorithm has the lowest blocking probability compare to the other two algorithms an the blocking probability increases smoothly as the request rate is getting higher. b. Noe failure Average hop count (b) ChinaNet (b) USANet Fig. Average hop count versus request rate with Blocking probability (b)chinanet (a) USANet (b) ChinaNet Fig. 7 Average hop count versus request rate with Figure 7 shows the average hop count of noe failure versus request rate with the ifferent numbers of channels with three ifferent algorithms in the USANet, ChinaNet, an NTTNet. The simulation results of single noe fault are similar to link fault for ifferent network topologies an it have more hop counts than the link fault scenario. When the request rate is higher an the channel number is smaller, the average hop count of recovery paths of three algorithms are all increase. It is notice that the average hop count of recovery paths of the three algorithms increases when the network topology is more complicate an has more recovery paths can be chosen. Blocking probability Figure 8 compares the blocking probability of noe ISBN: IMECS 9

6 Proceeings of the International MultiConference of Engineers an Computer Scientists 9 Vol I IMECS 9, March 18 -, 9, Hong Kong failure versus request rate with ifferent numbers of channels with three ifferent algorithms in the USANet, ChinaNet, an NTTNet. It is similar to the link fault scenario that when the request rate is high an the channel number is small, all the blocking probabilities of the three algorithms are increase. The propose algorithm has the lowest blocking probability compare to the other two algorithms an the blocking probability is increase smoothly as the request rate is getting higher. It also shows that the propose algorithm still has stronger availability to hanle noe failure situations an higher scalability than other two algorithms (a) USANet (b) ChinaNet Fig. 8 Blocking probability versus request rate with V. Conclusion In this paper, we have propose a ynamic core-base selection (DCS) algorithm for one-to-many multicast traffic with minimum spanning tree in WDM mesh networks. The propose algorithm can provie the local recovery for link failure or noe failure base on the information from core noes. The simulation results show that the propose algorithm outperforms the ual tree an MRLR algorithm in terms of the average hop count of recovery path an blocking probability, especially in the complicate network topology. The propose algorithm is scalable to large-size mesh networks an can achieve high network survivability an reliability. Furthermore, when multiple faults occur, the propose algorithm also can provie fast recovery ability an further research is the extension of more efficient core noes selection scheme to satisfy the many-to-many multicast services. We believe that the propose work will meet the requirements of future networks for high performance, scalability an reliability. References [1] B. Mukherjee, WDM optical communication networks: progress an challenges, IEEE Journal on Selecte Areas in Communications, Vol. 18, Issue, pp , Oct.. [2] A. Karaman an H. Hassanein, Core-selection algorithms in multicast routing - comparative an complexity analysis, Computer Communications, Vol. 29, Issue 8, pp , May 6. [3] D. Waitzman, C. Partrige, an S. Deering, Distance vector multicast routing protocol, IETF RFC 7, Nov [4] A. Aams, J. Nicholas, an W. Siaak, Protocol inepenent multicast - ense moe (PIM-DM): Protocol specification (Revise), IETF RFC 3973, Jan.. [] J. Moy, Multicast extensions to OSPF, IETF RFC 184, Mar [6] A.M. Al-Sharafi, M. Othman, M.N. Sulaiman, an S. Subramaniam, Moifying the multicast tree in AMTree protocol using ranom core selection, International Conference on Computer an Communication Engineering 8, Kuala Lumpur, pp , -1 May 8. [7] B. Fenner, M. Hanley, H. Holbrook, an I. Kouvelas, Protocol inepenent multicast - sparse moe (PIM-SM): Protocol specification (Revise), IETF RFC 41, Aug. 6. [8] A. Karaman an H. Hassanein, QoS-constraine core selection for group communication, Computer Communications, Vol., Issue 7, pp , May 7. [9] A. Chakrabarti an G. Manimaran, A case for tree migration an integrate tree maintenance in QoS multicasting, Computer Communications, Vol. 26, Issue 9, pp. 7-17, June 3. [] A. Fei, J. Cui, M. Gerla, an D. Cavenish, A ual-tree scheme for fault-tolerant multicast, IEEE International Conference on Communications, Helsinki, Finlan, Vol. 3, pp , June 1. [] I.S. Hwang, R.Y. Cheng, an W.D. Tseng, A novel ynamic multiple ring-base local restoration for point-to-multipoint multicast traffic in WDM mesh networks, Photonic Network Communications, Vol. 14, No. 1, pp , Aug. 7. [12] G. Maier, A. Pattavina, S.D. Patre, an M. Martinelli, Optical network survivability: protection techniques in the WDM layer, Photonic Network Communications, Vol. 4, No. 3, pp , 2. [] A.F. Elrefaie, Multiwavelength survivable ring network architectures, IEEE International Conference on Communications, Geneva, Vol. 3, pp , May 23-26, [14] J. Tapolcai, P.H. Ho; D. Verchere, T. Cinkler, an A. Haque, A new share segment protection metho for survivable networks with guarantee recovery time, IEEE Transactions on Reliability, Vol. 7, No. 2, pp , June 8. [1] M. Mear, S.G. Finn, R.A. Barry, an R.G. Gallager, Reunant trees for preplanne recovery in arbitrary vertex-reunant or ege-reunant graphs, IEEE/ACM Transactions on Networking, Vol. 7, Issue, pp , Oct [16] C. Boworntummarat, L. Wuttisittikulkij, an S. Segkhoontho, Light-tree base protection strategies for multicast traffic in transport WDM mesh networks with multi-fiber systems, IEEE International Conference on Communications, Paris, France, Vol. 3, pp , June 4. [17] L. Guo, H. Yu, an L. Li, A new share-path protection algorithm uner share-risk link group constraints for survivable WDM mesh networks, Optics Communications, Vol. 246, Issue 4-6, pp , Feb.. [18] D. Xu, Y. Xiong, an C. Qiao, Novel algorithms for share segment protection, IEEE Journal on Selecte Areas in Communications, Vol. 21, Issue 8, pp. -31, Aug. 3. [19] L. Luhua, L. Lemin, an W. Sheng, Multicast protection scheme in survivable WDM optical networks, Journal of Network an Computer Applications, Vol. 31, Issue 3, pp , Dec. 6. ISBN: IMECS 9