A Fast Recovery Technique for Multi-Point to Multi-Point MPLS tunnels

Transcription

1 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 34 A Fast Recovery Technique for Muti-Point to Muti-Point MPLS tunnes Mohaad Chaitou and Jean-Louis Le RouxOrange Labs avenue Pierre Marzin, 300 Lannion France Eais:Mohaad.Chaitou@gai.co, JeanLouis.Leroux@orange-ftgroup.co Abstract: This paper proposes fast reroute extensions to RSVP- TE in order to support the protection of utipoint to utipoint (MPMP) MPLS TE-tunnes. To protect an eeent (ink or node) of a priary MPMP TE-tunne, we propose to use an MPMP bypass TE-tunne connecting a set of nodes around the protected eeent. During faiure the priary MPMP TEtunne is encapsuated into the MPMP bypass TE-tunne which cas for defining a new type of MPLS hierarchy, the utipoint to utipoint hierarchy. The node of the priary MPMP TEtunne upstrea to the protected eeent, caed the Upstrea Protecting ode (UP), seects the MPMP bypass TE-tunne to be used for the protection. By extending the point to utipoint MPLS hierarchy, which reies on the upstrea abe assignent, we discuss severa extensions scenarios depending on the nuber of eaves of a bypass TE-tunne and on the nuber of UPs per bypass tunne. The scaabiity/bandwidth-consuption tradeoff between these schees is anayzed by eans of siuations. The proposed ethod can be efficienty used for the protection of point-to-point and point-to-utipoint TE-tunnes as we, as such tunnes are actuay particuar cases of MPMP TE-tunnes. I. ITRODUCTIO In [1], we have proposed extensions to the Resource ReSerVation Protoco-Traffic Engineering (RSVP-TE) for setting up Muti-Point to Muti-Point (MPMP) Muti Protoco Labe Switching-Traffic Engineering (MPLS- TE) tunnes. The ain otivation of this proposition is to reduce the nuber of TE-tunnes (a.k.a TE-LSP: TE- Labe Switched Path) needed to aintain utipoint connectivity between edge routers. Reducing the nuber of TE-LSPs aows reducing the nuber of contro pane and data pane states to be aintained on a node. A state denotes the data inforation that ust be stored at a node in order to aintain a TE-LSP. The nuber of states needed to aintain a TE-LSP on a router is at east equa to the nuber of neighbors of this router that are crossed by this TE-LSP. Reducing the nuber of TE-LSPs is of prie iportance as it iproves, in the contro pane, the scaabiity of RSVP-TE by decreasing the eory and CPU consued on a router. In addition, reducing the nuber of TE-LSPs reduces the ength of MPLS tabes in the data pane. The MPMP TE-LSPs ead to an iportant reduction in the nuber of TE-LSPs and hence in the nuber of states [1]. Additiona procedures are required to support Fast Reroute (FRR) for such MPMP TE-LSPs. This contribution defines such procedures. ote that the procedures defined in this paper can be used to protect the point-to-point and point-to-utipoint TE-LSPs as we because they are particuar cases of MPMP TE-LSPs. Without oss of generaity we consider in the foowing ony the protection of MPMP TE-LSPs. Mutiedia appications such as Vo or TV have strong resiiency requireents, with a target of sub-50s recovery upon faiure. Fast Re-Route (FRR) is a echanis used to protect MPLS-TE tunnes in the case of ink and/or node faiure. Traffic recovery is guaranteed within a 50 s tie interva foowing the faiure. To protect a priary PP TE-LSP, the point to point MPLS hierarchy is used []. That is, a bypass TE-LSP connecting the upstrea node of the protected eeent, caed Point of Loca Repair (PLR), to the downstrea node of the protected eeent, caed Merge Point (MP), is used to encapsuate the priary PP TE-LSP during faiure. To protect a priary PMP TE-LSP, the use of point to point hierarchy is proposed in [3] and the use of point to utipoint MPLS hierarchy is proposed in [4]. In this atter case the concept of Upstrea Labe Assignent (ULA) defined in [5], is used. Indeed, the defaut Downstrea Labe Assignent (DLA) ethod of MPLS ay ead to traffic dupication on the inks of the PMP bypass TE-LSP because distinct abe woud be aocated by distinct erge points. Due to the bidirectiona nature of an MPMP priary TE- LSP, its protection by using PP and /or PMP bypass TE-LSPs woud cause a scaabiity probe since a fu esh of PP or PMP bypass TE-LSPs woud have to be estabished between a nodes of the priary MPMP TE- LSP adjacent to the protected eeent. This woud aso require significant extensions to an RSVP-TE Path essage. Indeed, the atter woud have to incude path inforation reated to tree branches that are not downstrea. The scaabiity probe is caused by the fact that additiona bypass TE-LSPs ead to increasing the eory consued in a router in addition to the nuber of forwarding entries in an MPLS tabe as entioned before. To overcoe the above iitations this paper proposes to use an MPMP bypass TE-LSP to protect a ink and/or a node of a priary MPMP TE-LSP by defining the concept of MPMP MPLS hierarchy. Hence, one MPMP bypass TE-LSP ay encapsuate one or severa MPMP priary TE-LSPs, which is an iportant scaabiity iproveent. The node of a priary MPMP TE-LSP upstrea to the protected eeent, caed Upstrea Protecting ode (UP), seects the MPMP bypass TE- LSP. As severa priary MPMP TE-LSPs can be

2 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 35 encapsuated within the sae MPMP bypass TE-LSP, we ay have ore than one UP in an MPMP bypass TE-LSP. Hence, we define two odes for signaing MPMP bypass TE-LSPs: singe-up ode and uti- UP ode. In addition, the eaves of an MPMP bypass TE-LSP ay cover a neighbors of a protected eeent or just nodes of the priary MPMP TE-LSP adjacent to the protected eeent. A scaabiity/bandwidth tradeoff exists between these scenarios. Extensive siuations are used in order to evauate perforances of these odes and to copare the with the use of PP and PMP bypass TE- LSPs. This paper is organized as foows. Section II provides a genera overview of PMP hierarchy and MPMP TE- LSPs echaniss. Section III detais the procedures required to protect MPMP TE-LSPs with MPMP bypass TE-LSPs. This section presents aso siuation exapes. Finay, Section IV concudes the paper. II. BACKGROUD A. PMP MPLS Hierarchy This paper reies upon extensions to the PMP MPLS hierarchy in order to define the MPMP MPLS hierarchy. The reader is referred to [4], [5] for further detais about this subject. The PMP MPLS hierarchy can be used to encapsuate any PMP tunne into another PMP one. We briefy present this concept in the context of FRR, i.e., a PMP bypass TE-LSP connecting a PLR to one or severa MPs encapsuates a priary PMP TE-LSP. The PLR aocates the sae abe, caed backup abe or Upstrea Assigned abe (UA abe), that shoud be used by a the MPs, to identify the priary PMP TE-LSP during faiure. At each MP, this abe is instaed in an MPLS tabe caed context specific tabe, uniquey identifying the PLR. This is because a abe coision probe occurs in the case where two PLRs aocate the sae UA abe to the sae MP for two distinct LSPs and this abe is instaed in the defaut MPLS tabe on the MP. When an MP receives a packet, the outer abe of this packet shoud deterine the context specific tabe to be used for switching the inner abe (i.e. the UA abe). This phase is caed deterining the context of the UA abe. Generay speaking there are two odes: In soe cases the upstrea abe assigner can be deduced fro the utipoint interface on which the packet is received. This is the case for instance when the underying interface itsef is a PMP bypass TE-LSP here the abe assigner is the PLR, i.e. the PMP bypass TE-LSP root. The context is the outer abe of incoing packets received through this tunne. In other cases where the utipoint interface has utipe accesses (i.e. utipe upstrea abe assigners), such as a LA interface, or an MPMP LSP, the upstrea abe assigner can no onger be identified by the interface on which the packet is received. [5] proposes a soution to identify the upstrea LSR on LA interfaces. It consists in adding, before the upstrea assigned abe, a context abe that wi uniquey identify the upstrea LSR. The vaue of this context abe is inferred fro the v4 address of the upstrea LSR, that is, it is the owest 0 bits of the v4 address of the upstrea LSR. This ensures that the context abe is unique ony if the address ask on the LA is greater than 1 bits. B. MPMP TE-LSPs This section provides an overview of the signaing ethod of MPMP TE-LSPs as defined in [1]. An MPMP TE- LSP is a bidirectiona TE-LSP which connects a set of eaf nodes that can act indifferenty as sender or receiver. Each eaf has a bandwidth requireent for traffic it wi send, and these bandwidth requireents ay differ for distinct eaves. The signaing of an MPMP TE-LSP is initiated by a specific node caed the root node, that ay be a eaf or any transit node on the MPMP TE-LSP. An MPMP TE-LSP is expicity routed. The expicit route of an MPMP TE-LSP is a set of point-to-point paths fro the root to each eaf. That is, siiary to PMP TE-LSPs, an MPMP TE LSP is regarded as a cobination of pointto-point source-to-eaf (SL) sub-lsps fro the root towards each eaf where the ter "source" stands here for the root. On the one hand, the signaing ethod consists of sending Path essages fro the root towards the eaves. On the other hand, eaves send Resv essages towards the root. As ong as Path essages are forwarded through the expicit path, bandwidth is updated in order to account for bandwidth requireents of a eaves on both upstrea and downstrea direction. In addition a Path essage ust contain a abe caed upstrea abe to aow sending abeed packets on the upstrea direction. The downstrea direction is the forwarding direction of Path essages (fro the root to eaves) whie the upstrea direction is the forwarding direction of Resv essages (fro eaves to the root). Athough in the data pane the counications are bidirectiona, we define the reationship "node A is upstrea to node B" fro the view point of the contro pane. That is node A is upstrea to node B if node A sends Path essages to node B. The Resv essage ust contain at east a downstrea abe eeent in order to send packets in the downstrea direction as for PP and PMP TE-LSPs. To aow counication between a eaves on a transit node T, an upstrea abe advertised by node T to a downstrea node D is apped to the upstrea abe advertised by the upstrea node U to node T and the downstrea abes advertised by downstrea nodes other than D. Fig. 1 shows an exape of an MPMP TE-LSP. There are four eaves denoted as Li, i=1,,3,4. In a Path essage, we present the paraeters: Bw_U (upstrea bandwidth), Bw_D (downstrea bandwidth), U_Labe (upstrea abe) and sub-bws (bandwidth request for each eaf). In a Resv essage ony the Labe paraeter is shown because the other paraeters are the sae as those of the Path essage. There are four expicit routes signaed by the Root "R" in the Path essages: {R L1}, {R sr >L}, {R sr >L3} and {R sr >L4}. When forwarding the Path essage observe how Bw_U and Bw_D are updated. For instance, in the Path essage sent fro node "sr" to eaf "L", Bw_U is equa to the

3 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 36 bandwidth requested by "L", i.e. sub-bw, and Bw_D is the bandwidth su of a reaining eaves (i.e. L1, L3 and L4). An exape of data forwarding is the foowing: a packet received on interface "i5" with upstrea abe "8" is forwarded to the upstrea node "R" with the upstrea abe "7", and to downstrea nodes "L" and "L3" with abes "3" and "33" respectivey. Sub-Bw=1 L i8 Bw_U=1 Bw_D= =30 U Labe=30 Sub-Bw Fig. 1 Exape of an MPMP TE-LSP. III. THE PROPOSED FAST REROUTE METHOD FOR MPMP TE-LSPS A. Probes of PP and PMP Bypass TE-LSPs There are three probes in using PP and PMP bypass TE-LSPs to protect an eeent (ink or node) of an MPMP TE-LSP: 1-Scaabiity: to protect a node "X" of a priary MPMP TE-LSP "P" that has n neighbors which beong to "P" we need to rey on n ( n 1) unidirectiona PP bypass TE- LSPs or n PMP bypass TE-LSPs. - Significant extensions to RSVP-TE Path essages: the priary Path essage sent fro "X" to each of the n 1 downstrea nodes ust incude the identifiers of the n Sub-Bw3=5 i7 L1 R i1 i i3 i4 Path essage i1,,i10 interfaces Sub-Bw1=10 sr i5 i6 Bw_U=10 Bw_D=1+5+15=3 U Labe=6 Sub-Bw1 Bw_U=1+5+15=3 Bw_D=10 U Labe=7 Sub-Bw,3,4 i9 i10 Bw_U=5 L3 Bw_D= =37 U Labe=9 Sub-Bw3 Sub-Bw4=15 reaining downstrea nodes. This is required to seect PP and PMP bypass TE-LSPs and above a to signa the backup LSPs. The Path essages shoud be extended to support the indication of this inforation which represents a significant extension to RSVP-TE. 3-Data repication: this ay occur in the case of PP bypass TE-LSPs when two or severa bypass TE-LSPs having the sae ingress share the sae ink. B. Context of the Proposed Method with MPMP Bypass TE-LSPs L4 Bw_U=15 Bw_D=10+1+5=7 U Labe=8 Sub-Bw4 1) Suary of the Method A Priary MPMP TE-LSP is protected against ink and/or node faiure using an MPMP bypass TE-LSP. The MPMP bypass TE-LSP ay exacty cover the set of nodes on the priary LSP, neighbours to the protected eeent, or ay cover a the nodes neighbours to the protected eeent. In the forer case, caed the Exact Covering (EC) ode, traffic is ony sent to nodes on the protected MPMP LSP whie in the atter case caed the L Labe 3 i8 Labe 31 Labe 35 i7 L1 R L3 i1 i i3 i4 sr i5 i6 i10 Labe 33 Labe 34 i9 L4 Resv essage Fu Covering ode (FC) traffic is aso sent to nodes not on the protected MPMP LSP that wi drop the traffic. To protect a given priary MPMP TE-LSP, against the faiure of a given ink/node, we distinguish the upstrea neighbour caed UP and the downstrea neighbours caed Ps (Protecting odes). The Ps cover either a subset of MPMP bypass eaves (FC ode) or the fu set (EC ode). The UP is aso a eaf of the bypass TE-LSP. There is exacty one UP per protected LSP and protected eeent. The roe of the UP is to seect the MPMP bypass TE-LSP to be used to protect the priary MPMP TE-LSP and to aocate the backup abe that wi be used to identify the protected LSP during faiure. The UP ay be a eaf of severa existing bypass TE-LSPs in which case it has to seect one aong the to protect a new priary MPMP TE-LSP or it ay request the signaing of a new bypass TE-LSP. Such procedure is detaied atter in Section III.B.). After seecting an appropriate bypass TE-LSP, the UP aocates a backup abe that wi be used during faiure to identify to which protected LSP the traffic encapsuated within the bypass TE-LSP beongs. This abe is then counicated to each P in a Path essage that aso indicates the bypass TE- LSP, and optionay the abe identifying the UP. These Path essages are caed backup Path essages. The sae backup abe is used in both directions between the UP and Ps. It is aocated by the UP to a Ps using upstrea abe assignent procedures. As discussed earier, upstrea abe assignent aows having the sae abe aocated between a set of nodes but requires on receivers the identification of the abe assigner, to avoid any abe coision between two assigners. Here the abe assigner is the UP. Hence a node neighbour to a protected eeent wi aintain one context specific MPLS tabe per UP. A node that receives a packet with the backup abe wi forward it in the context of the UP that aocated it. This context wi be identified on the receiving node either ipicity fro the MPMP bypass TE-LSP on which the packet is received or expicity using an additiona abe caed Context abe (C-abe) that uniquey identifies an upstrea abe assigner. In the forer ode caed Ipicit Identification ode (I-I ode) there ust be exacty one UP per bypass TE-LSP (singe-up ode) because the bypass TE-LSP is used to uniquey identify the UP. In the atter case caed Expicit Identification (E-I ode) there ay be severa UPs per bypass TE- LSP (uti-up ode), i.e. severa upstrea abe assigners. In order to guarantee the unicity of the context abes (C-abes) we cannot rey on the ethod used for LA interfaces, i.e., on inferring the C-abe fro the v4 address of the UP. We propose to assign the C- abes offine by a server where each C-abe corresponds to a unique node in the network. A range of C-abes ust be reserved, i.e., a node shoud not aocate a abe within this range. During node faiure, the traffic is tunneed by the UP and Ps within the MPMP bypass TE-LSP towards a other Ps. The outer abe is the MPMP bypass abe used to transport traffic between the UP and Ps, avoiding the faied eeent. The inner abe is

4 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 37 the backup LSP abe that is used to identify the protected MPMP TE-LSP and redirect the traffic to this LSP on the UPs and Ps. There are two iportant procedures in the proposed ethod: the seection of the bypass TE-LSP by a UP and the procedures of the UP and Ps before and during faiure. ) Seection of Bypass TE-LSPs a) Signaing the Bypass TE-LSP An MPMP bypass TE-LSP is signaed by its root. As entioned in [1], the choice of the root can be constrained by severa optiization criterions such as iniizing the nuber of essages used to setup the MPMP bypass TE-LSP. The paraeters of the MPMP bypass TE-LSPs ust be configured on its root. Such paraeters incude: the protected eeent, the type of the bypass TE-LSP (FC or EC), its ode (singe-up, uti-up), the set of eaves of this bypass TE-LSP and their required bandwidths. The choice of the root and the configuration of the bypass TE-LSP paraeters can be perfored offine by a network anageent syste (MS) or onine by the UP, acting itsef as root LSR. In this atter onine case, soe paraeters of the MPMP bypass TE-LSP, naey its type (EC/FC), its ode (singe-up/uti-up) and the bandwidths of the eaves can be ocay configured on the UP. The set of eaves of the bypass TE-LSP is deterined by the UP as foows. For an FC-bypass, the set of eaves is deterined easiy by finding a the neighbours of the protected eeent in the IGP database. For an EC-bypass, the set of eaves is deterined by inspecting the <ERO> object of the Path essage of the priary TE-LSP; the eaves are then the UP itsef and a downstrea nodes of the protected eeent, caed Ps. After choosing the root and configuring the paraeters of the MPMP bypass TE-LSP on it, the root signas the MPMP bypass TE-LSPs by using the procedures defined in [1], i.e. by sending Path essages identifying the MPMP bypass TE-LSP to the eaves. We propose to extend these Path essages as described in Section III.B.4) in order to counicate the protected eeent, the type (EC/FC) and the ode (singe/uti UP) of the bypass TE-LSP to the eaves. b) Seecting the Bypass TE-LSP On receipt of a Path essage for a priary TE-LSP, a UP has to seect a bypass TE-LSP aong a set of bypass TE-LSPs for which it is a eaf. The seected bypass TE- LSP ust cover at east a the Ps of the protected LSP. This requires that the UP knows the eaves of the candidate bypass TE-LSPs. The UP wi know the eaves of a bypass TE-LSP in two cases: 1- An exact covering MPMP bypass TE-LSP for which it is the root. In this case the UP knows the eaves because it is the root of the bypass TE-LSP. - A fu covering MPMP bypass TE-LSP. Here the eaves are a the neighbours to the protected eeent. These neighbours can be found in the IGP database. In return note that a UP cannot know the eaves of an EC bypass for which it is not the root. This woud require drastic and undesired extensions to RSVP-TE, indeed this woud require aways indicating a the bypass eaves in a Path essage signaing a bypass TE-LSP. Fro 1 and we deduce that an EC bypass ust be singe UP, and the UP ust be the bypass TE-LSP root. In Tabe 1 we propose the situations in which a bypass TE-LSP of a given type and ode can be seected by a eaf acting as a UP and which identification ode can be used in each situation. Tabe 1. Bypass Types and Identification Modes Singe UP Mutipe UPs EC-bypass OK (E-I ode, I-I ode) OK FC-bypass OK (E-I ode, I-I ode) OK (E-I ode ony) ote that there are two restrictions in the third coun of Tabe 1. First, the E-I ode ust be used with the uti- UP ode. As entioned before, this is because there are severa UPs in the bypass TE-LSP and hence the atter cannot be used to uniquey identifying each UP separatey. Second, for the EC-bypass we aow ony the singe-up ode for the reasons expained before. The bypass TE-LSP seection procedure by a UP, for a given protected LSP becoes as foows. The candidate bypass TE-LSPs for the UP wi be: -EC bypass LSPs for which it is the root, and that exacty cover the set of Ps. -Singe UP FC bypass LSPs for which it is the UP. -Muti UP FC bypass LSPs. Then the detaied seection procedure aong these candidates is driven by oca poicy decision on the UP. 3) Procedures Before and During Faiure To signa a backup LSP, after having seected an MPMP bypass TE-LSP, the UP sends one Path essage towards each P, indicating the backup abe, which is upstrea assigned, and its context. A P node instas this upstrea assigned abe in the context specific MPLS tabe corresponding to the UP; it is apped towards downstrea interfaces and abes of the protected LSP. Hence, a backup abe wi be instaed in the MPLS tabe corresponding to its UP node, identified either ipicity by the bypass TE-LSP on which it is received or expicity by the context abe. Ps aso insta a backup output for the upstrea direction. Upon faiure the traffic wi be redirected within the MPMP bypass using the backup abe as inner abe. Siiary the UP wi insta the backup abe it assigned in a specific MPLS tabe corresponding to itsef as UP; it is apped to the upstrea interface and abe as we as other downstrea interfaces and abes. Aso it instas a backup output for the downstrea direction, and upon faiure traffic wi be redirected within the MPMP bypass using the backup abe it assigned as inner abe.

5 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 38 4) Extensions to RSVP-TE The foowing extensions are proposed to RSVP-TE Path essage: 1- Two fags shoud be added; one to specify the type of the bypass TE-LSP (EC/FC) and another one to specify the ode (singe-up, uti- UP). These fags can be added in the <LSP_Attributes> object [7]. - An object shoud be added to the Path essage. This object indicates the identity of the eeent protected by the bypass TE-LSP (a ink or node). Ca this object the <Protected Eeent> object. 3- A new TLV which contains the C-abe ust be supported by the <IF_ID RSVP_HOP> object [6]. The fags and the <Protected Eeent> object are required ony in a Path essage signaing a bypass TE- LSP. Hence, their presence in a Path essage indicates that the signaed TE-LSP is a bypass TE-LSP. The new <IF_ID RSVP_HOP> TLV wi be required in the Path essage that signas the backup abe for a given protected LSP, in order to indicate the context abe. 5) An exape Fig. shows step by step how an MPMP bypass TE-LSP is signaed. This bypass TE-LSP caed "B" has as root and, and as eaves. It is an FC-bypass aiing to protect with uti-up ode. During the signaing phase (Fig. (a) and Fig. (b) a eaves of "B" wi be aware about the eeent protected by this bypass TE-LSP, i.e., its type (i.e. FC) and its ode (uti- UP). Beyond this inforation the signaing phase is exacty the sae as that of MPMP TE-LSPs (Section II.B). At the end of this step, the MPLS tabes of the nodes crossed by "B" are popuated. As shown in Fig. (c), there are two types of MPLS tabes at a eaf of the bypass TE-LSP "Ri"i=1,6,7: the ain MPLS tabe caed (Ri, M) and the context specific tabes caed (Ri, Rj). A tabe (Ri, Rj) represents the context specific tabe of node Rj instaed at node Ri. ote that since the ode of the bypass TE-LSP is FC each eaf of "B" wi find a the neighbours of by using its IGP database and wi then create an MPLS context specific tabe for each such neighbour if such tabe did not exist. An exape of packet forwarding is given in Fig. (d) and Fig. (e) respectivey. ext in Fig. 3 we show a priary MPMP TE-LSP, caed LSP1, which crosses,,,,,,,, 0 such that is the root of LSP1 and is the upstrea of and according to the notion of upstrea expained in Section II.B. We show in Fig. 3(a) the MPLS tabes after signaing LSP1. Suppose that the protected node is. Hence within LSP1, is the UP of. Thus, searches for a bypass TE-LSP that can be used to protect against faiure of. It founds that the bypass TE-LSP "B" shown in Fig. can be used to protect LSP1 because the nodes downstrea to within LSP1, i.e. and are eaves of "B". and are caed the Ps within LSP1. The UP wi then te and that they shoud use "B" to encapsuate LSP1 during faiure of. This is shown in Fig. 3(b) by two Path essages sent directy fro to and. This phase is caed the backup signaing phase in which aocates the upstrea assigned abe and specifies its context. We suppose the expicit identification ode is used (required because "B" is uti-up bypass). Hence, the Path essage carries a context abe identifying the UP (Fig. 3(b)). This context abe wi be used by and to identify in the data pane the assigner of the upstrea assigned abe that is. The context abe identifying (95) is instaed in the ain MPLS tabes (,M), (,M) and (, M) and redirected to the corresponding -context MPLS tabe, i.e. (,), (,) and (,) respectivey (Fig. 3(c)). The context MPLS tabes (,), (,) and (,) wi contain an entry apping the upstrea assigned abe () to LSP1 downstrea segents (Fig. 3(c)). In addition, backup outputs wi be added to the ain MPLS tabes in this phase (Fig. 3(c)). These outputs are used ony when fais to repace the outputs eading to in the MPLS tabe (Fig. 3(c)). Such outputs encapsuate the traffic of the priary LSP, LSP1, into the bypass TE- LSP "B". As shown in Fig. 3(c) there is a stack of three abes: the outer abe is the abe of "B", the second abe is the C-abe (95) and the inner abe is the upstrea assigned abe (). Fig. 3(d) and Fig. 3(e) show respectivey an exape of packet forwarding before and after faiure of. and wi be abe to forward the incoing packets received through "B" to the LSP1 segents by consuting their MPLS tabes suarized in Fig. 3(c). Path to Protected eeent= Type=FC Mode=Muti-UP Upstrea abe= 5.a- Signaing the bypass TE-LSP..b-Signaing the bypass (Cont.) Resv to abe=55 etc Path to Protected eeent= Type=FC Mode=Muti-UP Upstrea abe=30 Path to : Protected eeent= Type=FC Mode=Muti-UP Upstrea abe=40 Resv to abe=50 etc Resv to abe=4 etc Path to Protected eeent= Type=FC Mode=Muti-UP Upstrea abe=35 Resv to abe=45 etc 0 0

6 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 39 MPLS tabe (, M) Three context tabes at : (,), (,), (,) Packets 5, 55 pop 55 Three context tabes at : (,), (,), (,) Packets 40, 4 pop 4 0 MPLS tabe (,M) Path essages to and Sender= Backup LSP1 identifier UA abe1= Context= C-abe =95 Seected Bypass: «B» o abe request MPLS tabe (,M) 5 50, 30 55, MPLS tabe (,M) 50 4, 45, 35 30, 4, 40 30, 45, Three context tabes at : (,), (,), (,) MPLS tabe (,M) Packets 35, 45 pop 45 UP of LSP1= P1 of LSP1= P of LSP1= 0.c- MPLS tabes Begin here: a packet arrives at d- exape1: a packet arrives at b-Backup signaing phase MPLS tabe (,) 68, 6, MPLS tabe (,M) Packets 5, 55 pop , 64, 66 68, 64, 70 68, 6, 95 pop 95, go to (,) Backup outputs (during faiure of ): 60 [5;95;], 66 [5;95;], 3.c- MPLS tabes after the backup signaisation phase. Begin here: Two MPLS packets arrive at : -One fro (abe 66) -another one fro (abe 60) MPLS tabe (,) 80, 8,0 MPLS tabe (,M) Packets 40, 4 pop , 8, , 8, , 80, 95 pop 95, go to (,) Backup outputs (during faiure of ): 84 [40;95;], 86 [40;95;], MPLS tabe (,) 88, MPLS tabe (,M) Packets 35, 45 pop , 90 78, 95 pop 95, go to (,) Backup outputs (during faiure of ): 90 [35;95;], Begin here: a packet arrives at.e- exape:a packet arrives at Fig. A bypass TE-LSP signaing exape MPLS tabe (,M) Packets 5, 55 pop , 64, 66 68, 64, 70 68, 6, 64 7, 74, 76 70, 74, 78 70, 7, MPLS tabe (,M) MPLS tabe (,M) Packets 40, 4 pop , 8, , 8, , 80, 0 3.d- Packet forwarding before faiure fais MPLS tabe (,M) 4 3.a- MPLS tabes after signaing LSP1. Packets 35, 45 pop , 90 78, e- packet forwarding during faiure

7 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 40 Fig. 3 FC-bypass TE-LSP with Muti-UP and expicit identification odes. C. Perforance Evauation In this section we evauate the use of MPMP bypass TE- LSPs in ters of scaabiity and bandwidth wastage by using a siuator written in C. We consider a scenario depicted in Fig. 4. ode "P1" has, neighbors. There are, 1 priary TE- LSPs which cross "P1". We assue that these LSPs have the sae characteristics (e.g. the sae bandwidth requireents). Each of those priary LSPs crosses the sae nuber of nodes, aong the neighbors of "P1" but not necessariy the sae set of neighbors. This assuption (i.e. the nuber is the sae for a priary LSPs) is justified because we intend to copare severa bypass TE-LSP scenarios under the sae conditions. The set of neighbors crossed by a priary LSP is randoy unifory distributed aong the neighbors of "P1". Iportant Reark: The MPMP bypass TE-LSP can be used to protect PP (=), PMP and MPMP TE-LSPs (>=). In Fig. 4 ony one priary MPMP LSP, caed LSP1, is shown. To protect "P1" we consider that the bypass TE- LSPs are signaed such that they cross a network connected to the neighbors of "P1" without using any ink between "P1" and any of its neighbors. Without oss of generaity, we can consider that this network is reduced to one node "P" as iustrated in Fig. 4. In this figure we show the case where a singe MPMP bypass TE-LSP exists; this is the case of an FC-bypass TE-LSP with uti-up ode. It can be observed fro Fig. 4 that the bandwidth wastage occurs at nodes that do not beong to LSP1 such as node in Fig. 4. In a uti-up bypass TE-LSP such as the one iustrated in Fig. 4 the bypass TE- LSP can be used to encapsuate a the priary LSPs crossing a subset of the neighbors of "P1". In this case the bandwidth oss is equa to priary LSP crosses LSP has ( )/ because each nodes whie an FC-bypass TE- eaves and hence there are ) ( nodes that ust drop traffic during faiure of "P1". If the bypass TE-LSP shown in Fig. 4 is singe-up, then the priary LSPs which can be encapsuated by this bypass TE-LSP ust cross the node "". In addition, node "" ust be the upstrea node of "P1" in a these priary LSPs, that is, "" is the singe UP (see Section II.B for the definition of the upstrea node in the case of MPMP LSPs). The bandwidth oss in this case is the sae as that of the uti-up ode. In an EC-bypass the bypass TE-LSP ust connect ony the neighbors of "P1" that beong to a priary LSP. For instance in Fig. 4 such a bypass TE-LSP woud have connected nodes "1", "",, "" so bandwidth wastage is avoided in this case. If another priary LSP, say LSP, crosses the sae set of neighbors of "P1" as that crossed by LSP1, then LSP can be encapsuated into this bypass TE-LSP if and ony if node "" is the UP in LSP. This is because we have defined ony the singe- UP ode for an EC-bypass (see Tabe 1). If the protection by PP bypass TE-LSPs is desired then we consider a fu esh of bypass TE-LSPs between the neighbors of "P1" crossed by a priary LSP. If each neighbor of "P1" is crossed by at east one priary LSP then we wi have ( 1) bypass TE-LSPs. To define the bandwidth oss, observe that in the particuar case of Fig. 4, if PP bypass TE-LSPs are used to protect LSP1 there wi be ( 1) tunnes started fro "1". A these tunnes use the ink "1 P". By observing that a these bypass TE-LSPs transport the sae inforation, one ay concude that the proportion of data dupication on ink "1 P" is ( )/( 1). We define the bandwidth oss in this case as the data dupication proportion on such inks. The bandwidth oss wi occur on such a ink if the neighbor which is connected to "P" beongs to at east one priary LSP. Siiary for PMP bypass TE-LSPs, we consider a fu esh of bypass TE-LSPs between the neighbors of "P1" crossed by a priary LSP. For instance, in order to protect "P1" in Fig. 4, we wi have a tota of bypass TE-LSPs. If another priary LSP, say LSP, crosses the sae set of neighbors of "P1" as that crossed by LSP1, then the PMP bypass TE-LSPs used to protect LSP1 can be re-used to protect LSP. It can be observed that there is no bandwidth oss in this case. In addition to the bandwidth oss defined above for the severa scenarios, we ai then to copute the nuber of states caused by signaing the bypass TE-LSPs through "P", i.e. the nuber of states in "P". ote that in the genera case, these states wi not necessariy be ocated on the sae node; they ay be spread aong severa nodes of the network. As entioned before, the scaabiity denotes the nuber of TE-LSP states. For each segent of the TE-LSP crossing the node we ust add a state. For instance, in Fig. 4 LSP1 generates states at "P1", whie the bypass TE-LSP in the sae figure generates states at "P". A PMP bypass TE-LSP crossing neighbors of a node and the node itsef wi generate states at this node whie a PP bypass TE-LSP wi generate two states at a crossed node. Fig. 5 shows the ipact of the nuber of priary LSPs on the nuber of states and on the bandwidth oss. The eft side coun of sub-figures represents the case where "P1" is a core node that is not connected to edge nodes.

8 Bandwidth oss (%) Bandwidth oss (%) uber of states at "P" uber of states at "P" M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 41 During faiure drops a packets received through the Bypass tunne because it does not beong to LSP1 Priary LSP1 A fu covering Bypass tunne Fig. 4 Scenario used for siuations. In this case "P1" has a ow degree so we suppose that 5, 3. The right side coun of sub-figures depicts the case where "P1" is connected to edge nodes. In this case "P1" has a high degree so we suppose that 0, 15. Indeed, operators increase the nuber of edge nodes in order to face the increase in the nuber of their cients which expains the high degree of core nodes connected to edge nodes. It can be shown fro Fig. 5 that the use of MPMP bypass TE-LSPs highy iproves the scaabiity by eading to an iportant decrease in the nuber of states Fig. 5 uerica resuts. This better scaabiity copared to the use of PP and PMP bypass TE-LSPs can be expained by observing that the nuber of states in PP case is a function of n where n and in PMP case is of the order whie it is proportiona to or for the three MPMP bypass TE-LSP scenarios. For the PP case, if n nodes, such that n, are traversed by at east one priary LSP then the nuber of states wi be n ( n 1). ote that as "" increases, n first increases and then becoes equa to. This expains the behavior of the nuber of states for PP bypass TE- LSPs shown in Fig. 5. For the PMP case, to protect a nuber of priary LSPs crossing the sae set of nodes we need 1= P of LSP1 However for two particuar reaistic cases we have shown that this bandwidth oss is ess than that of PP bypass TE-LSPs. The FC-bypass singe-up eads to a nuber of states cose to that of the EC-bypass case. However, the EC-bypass case does not introduce a bandwidth oss whie in the case of the FC-bypass singe UP, a non negigibe bandwidth oss is observed. Finay, the PMP bypass TEnetwork UP of LSP1 P1, <= uber of priary LSPs crossing "P1" PP PMP FC singe UP FC uti UP EC =5, =3 =5, =3 PP MPMP (FC Bypass tunnes) uber of priary LSPs crossing "P1" = P of LSP PP PMP FC singe UP FC uti UP EC, <= PMP bypass TE-LSP and each bypass 1 P =0, =15 P uber of priary LSPs crossing "P1" =0, =15 PP MPMP (FC Bypass tunnes) uber of priary LSPs crossing "P1" TE-LSP generates states as entioned before. Observe that there are at axiu PMP bypass TE-LSPs and hence states; this depicts the case where each priary LSP crosses a set of nodes such that this sae set is not crossed by the other priary LSPs. On the other hand, the nuber of states in the case of an FC-bypass TE-LSP with uti-up ode is sipy. For an FC-bypass TE-LSP with the singe UP ode wi have at axiu in(, ) bypass TE-LSPs. In this case there wi be in(, ) states. Siiary, for an EC-bypass TE-LSP the axiu nuber of bypass TE-LSPs is. In this case the nuber of states is. That is the nuber of states in the case of the three MPMP scenarios increases ineary with respect to or whie it is a function of for PP and PMP bypass TE-LSPs which expains the better perforance observed in the case of MPMP scenarios. The gain in scaabiity for the MPMP scenarios is atered by a bandwidth oss in the case of using FC-bypass TE-LSPs (singe or uti-up odes). The second row of subfigures in Fig. 5 shows that for the two particuar cases 5; 0 the traffic dupication observed in the case of PP bypass TE-LSPs is greater than the bandwidth oss of the MPMP scenarios for 1. ote that intuitivey, there is no bandwidth oss in the case of MPMP EC scenario. Fig. 5 shows aso that the nuber of states for the two cases EC-bypass and FC-bypass singe UP are very cose. This can be expained by observing that on one hand the nuber of bypass TE-LSPs in the case of the FC-bypass singe UP shoud be ess than that of the EC-bypass case. This is because two priary LSPs use the sae bypass TE-LSP if: 1) they have the sae UP in the case of the FC-bypass singe UP. ) They have the sae set of "" nodes in addition to the sae UP in the case of the EC-bypass. On the other hand, a bypass TE-LSP generates "" states in the case of the FC-bypass singe UP and "" states in the case of the EC-bypass. We concude that as the degree of a core node increases the use of MPMP bypass TE-LSPs becoes ore and ore interesting in ters of scaabiity (this is confired by ore siuations that are not shown due to the ack of space). In particuar, the EC-bypass case, which does not ead to bandwidth oss, has shown good perforance aso in ters of scaabiity. The FC-bypass uti-up case is the best in regards of enhancing the scaabiity but it eads to bandwidth oss if.

9 M. Chaitou and J. L. Roux / IJECCT 01, Vo. (3) 4 LSPs case has the worst perforance in ters of scaabiity because the nuber of states is of the order of. The reduction in the nuber states is caused by a reduction in the nuber of bypass TE-LSPs, i.e. scaabiity in both contro and data pane is enhanced by using MPMP bypass TE-LSPs. IV. COCLUSIO In this paper we focused on protection in utipoint to utipoint MPLS TE-LSPs. We showed that existing PP and PMP Fast Reroute echaniss are not suited to MPMP TE-LSP and proposed a new ethod reying on using an MPMP bypass TE-LSP for ink and node protection. During faiure traffic on a protected MPMP TE-LSP is tunneed towards the nodes adjacent to the faied eeent, within an MPMP bypass TE-LSP avoiding the faied eeent. Leaves of the bypass TE-LSP use the sae upstrea assigned abe to identify the protected LSP encapsuated within the bypass TE-LSP, and this aows avoiding data dupication. This abe is signaed by the node upstrea to the protected eeent using RSVP-TE upstrea abe assignent procedures. We have discussed severa scenarios depending on the coverage of bypass TE-LSP eaves and the nuber of upstrea nodes that can use the sae MPMP bypass. We showed that using MPMP bypass TE-LSPs is better in ters of scaabiity than using PP and PMP bypass TE-LSPs. Aso, we showed that a tension between contro pane scaabiity and data pane bandwidth savings exists and we highighted that the exact covering MPMP bypass ode provides a good tradeoff. Future investigations wi focus on extending the proposed soutions to other switching techniques such as ATM, switched Ethernet, etc. REFERECES [1] Chaitou, M., Le Roux, J.L.: Enhancing the scaabiity of Traffic Engineering Labe switched Paths. Subitted to IJECCT. [] Pan, P., Swaow, G., Atas, A.: Fast Reroute Extensions to RSVP- TE for LSP Tunnes. RFC 4090 May [3] Aggarwa, R., Papadiitriou, D., Yasukawa, S.: Extensions to RSVP-TE for Point-to-Mutipoint TE-LSPs. RFC 4875, Jan [4] Le Roux, J-L.: PMP MPLS-TE Fast Reroute with PMP bypass Tunnes. Work in progress, May 007, draft-ietf-ps-pp-tebypass-00.txt [5] Aggarwa, R.: MPLS Upstrea Labe Assignent and Context Specific Labe Space. Work in progress, Mar. 007, draft-ietfps-upstrea-abe-0.txt [6] Aggarwa, R., LeRoux, J-L.: MPLS Upstrea Labe Assignent for RSVP-TE. Work in progress, Mar. 007, draft-ietf-ps-rsvpupstrea-01.txt [7] Farre, A., et a: Encoding of Attributes for MPLS Labe Switched Path (LSP) Estabishent Using RSVP-TE. RFC 440, Feb