MultiProtocol Label Switching - MPLS ( RFC 3031 )

Transcription

1 Outline MultiProtocol Label Switching - MPLS ( RFC 3031 ) 1. What is MPLS and how does it work? 2. What MPLS is used for? 3. Label Distribution Protocols 1

2 1. What is MPLS and how does it work? MPLS is a technology which enables to forward L3 packets according to labels at L3. This is in contrast to ordinary L3 forwarding which is based on e.g. the IP DA and shortest path routing in TCP/IP. Multiprotocol because the techniques presented here are applicable to ANY network layer protocol, not just IP. LSR MPLS domain Ingress LSR LSP edge LSR Egress LSR 2

3 Terms : Label Switching Router - LSR : a device that forwards IP packets by label switching, aware to MPLS control protocols and operates one or more network layer routing protocols. MPLS domain : a group of connected LSRs. Label Switched Path - LSP : a route of a packet. Forwarding : the operation in which a router/switch receives a packet on an incoming interface and decides to which output interface to forward the packet, based on fields in the packet ( header ). Switching : forwarding packets based on local information in the device, e.g. a label. Routing : forwarding packets based on global information, e.g. IP addresses. 3

4 Forwarding Equivalence Class - FEC : all the packets that are forwarded by an LSR by the same way, i.e. forwarded to the same outgoing link with the same label. E.g. all packets that are destined to the same egress LSR ( of the MPLS domain ). Label Distribution Protocol - LDP : A protocol for the establishment of LSPs. It is a set of procedures by which one LSR informs another LSR of the (FEC,label) bindings it has made. 4

5 The label swapping concept LSR Arriving packet Old label Forwarding Table Incoming old Outgoing new next link label link label hop New label Label swapping This is the IP address of the next LSR on the LSP. The IP address might sometimes be mapped to a physical address (ARP). 5

6 An LSR implements routing protocols. However, it forwards packets based on a label and not on a Destination Address. Control Plane : Routing protocols, e.g. OSPF, PNNI Label Distribution Protocols Data Plane : Switching hardware and software 6

7 Several tasks need to be solved : 1. How LSPs are generated, i.e. how LSRs decide and notify their neighboring LSRs along a LSP about their labels? When forwarding packets : 2. How a label is transmitted from an LSR to the next one along a LSP? ( in a layer 2 header, layer 3 header? ) 3. How does a LSR know which is the next LSR along a LSP? 7

8 Task 2: The label is carried as either a part of the L2 header, if the L2 has a label defined in the header, e.g. VCI/VPI in ATM, DLCI in Frame Relay, VLANs in Ethernet etc. Or, it is added between the Network layer header ( IP ) and the link layer header. Such an header is called a Shim Header. Payload Network Shim Link layer layer Header Header Header L3 L2 8

9 LSRs can be connected by different L2 technologies. An LSR decides on an outgoing link and a label and constructs a packet according to the L2 technology. E.g IP over Ethernet : Ethernet Frame IP packet Label is here Payload IP Shim Ethernet Header Header Header LSR Ethernet LSR 9

10 Or: IP packet Ethernet Frame Ethernet header Payload IP Header VLAN ID SA DA Label is here LSR ATM and Ethernet : Ethernet LSR LSP LSR ATM LSR Ethernet LSR VPI/VCI are used as the MPLS label Shim Header or VLAN ID are used as the MPLS label 10

11 Task 3: An entry in the Forwarding Table looks as follows : Incoming interface Outgoing Interface Incoming label Outgoing Label Next Hop Address The next hop address is the IP address of the next LSR on the LSP. E.g. over an Ethernet subnet the given next hop address chooses among several LSRs that might be connected to the subnet. Notice that it is possible that the LSR will use ARP, in case of IP, to resolve the PHY address of the next hop LSR. 11

12 LSR IP address of the next LSR is mapped to a physical address. LSR LSR Task 1: By Label Distribution Protocols, to be described below. 12

13 2. What MPLS is used for? * Unified forwarding algorithm - looking for a label in a table. This can work for both unicast and multicast. Currently, the forwarding for unicast ( by DA ) is different from that for multicast ( SA, DA ). Unified Forwarding component Unicast Unicast routing Multicast routing with TOS routing Forwarding algorithm in conventional routing Longest match on DA Longest match on DA + exact match on TOS Exact match on SA, DA, incoming interface Forwarding algorithm in label swapping Label swapping 13

14 * Enabling routing according to several fields, not just the DA. E.g. based on the interface on which the packet arrived, based on several fields in the packet, etc. A packet is assigned once to a FEC by the ingress LSR. Other LSRs only forward by the label. * The assignment to a FEC by the ingress LSR reduce processing overhead from internal routers. This has importance, e.g. in core networks where router performance is critical. * Traffic Engineering - packets can be forwarded according to LSPs and not according to the ordinary IP routing. This enables to use the network resources more efficiently because IP routing is based on shortest paths which use only a part of the network resources. * QoS routing - packets are forwarded through LSPs with guaranteed QoS performance. This is achieved by allocating resources to LSPs. 14

15 * Routing scalability improvement Internal routers in a Transit AS need to keep a list of exterior destinations, in order to route transit traffic to the correct border routers. If instead, transit traffic is forwarded by LSPs, from ingress to egress border routers, the internal routers can keep only labels, instead of a list of external destinations. A LSR which is also a BGP router Transit AS LSP LSP A BGP router maintains a LSP with every other BGP router to which it routes transit traffic. 15

16 * Enabling a simpler scheme for IP traffic over ATM Overlay model The ATM network is another L2 technology and as such its control and management infrastructure are separate from that of the IP routers. Classical IP over ATM : routers communicate with PVC/SVC through the ATM network router host ATM switch ATM network 16

17 Problems 1. The system manager needs to manage two protocol architectures. 2. Any router has (n-1) neighbors in the IP level. This imposes a lot of overhead at the routers if n is large. 3. A topology change, e.g. a link failure, can interrupt the communication between several pairs of routers connected to the network. This can lead to many routing updates because o(n) links between routers failed. 4. There is a need to adjust IP QoS demands to those of ATM. 5. ARP is needed to translate between IP and ATM addresses. 17

18 A better approach - the Integrated Model. The ATM switches are turned to be LSRs which run IP control protocols, such as IP routing protocols, and MPLS control protocols, e.g. LDP protocols. They communicate by ATM cells. Thus: Control plane - TCP/IP & MPLS Data plane - ATM Main advantages : 1. A system manager needs to maintain only one protocol architecture - TCP/IP. 2. Less overhead because the IP routers only consider as neighbors those ATM switches to which they are physically connected. 18

19 An LSP ( ATM PVC/SVC ) is established between any two routers. IP packets are segmented into ATM cells by the router/ingress LSR. Router+LSR host Router+LSR (ATM hardware) peers 19

20 ATM ARP IP MARS NHRP PNNI Q.2931 ATM Hardware IP Routing protocols Binding and LDP protocols ATM Hardware 20

21 3. Label Distribution Protocols - LDP The control plane 1. Routing protocols. 2. Algorithms for FEC definition and label allocation. 3. Label Distribution Protocols. Network layer routing protocols ( e.g., OSPF, BGP, PIM ) Procedures for the creation of bindings between labels and FECs Procedures for distributing information about created label bindings - Label Distribution Protocols. Maintenance of Forwarding Table 21

22 Routing protocols : enable to decide on the route of an LSP, either based on shortest path, or enables the computation of explicit routes. Local procedures decide of FECs and allocate labels. Label Distribution Protocols distribute binding of FECs and labels. Three main LDPs are defined : LDP based on hop-by-hop routing (every LSR decides on its next hop independently) and a FEC is an Address Prefix. RSVP-TE An extension to RSVP that is used to support explicit routing with or without resource reservation ( QoS ). CR-LDP An extension to LDP that is used to support explicit routing and resource reservations ( QoS ). 22

23 Some terminology related to LSPs LSR2 is downstream in relation to LSR1 if they are a part of an LSP and data packets in the LSP are transmitted from LSR1 to LSR2. LSR1 Data packets LSR2 Label Distribution Downstream on Demand : LSR1 asks LSR2 for a (FEC,label) binding. UnSolicited Downstream LSR2 sends a (FEC,label) binding to LSR1 without an explicit request. 23

24 Liberal vs. Conservative label retention Liberal - An LSR keeps a label from a neighbor LSR for a FEC even when the neighboring LSR is not the next hop for the FEC. Conservative - An LSR does not keep a label when the label is not needed. The advantage of the Liberal approach - there is a label immediately when one is needed. e.g. when the LSR becomes the next hop for the FEC. The disadvantage - labels are wasted because an LSR distributes unnecessary labels. 24

25 Control modes : Ordered vs. Independent Next Hop (for FEC) Definition Incoming Label Independent LSP control * Each LSR makes independent decision on when to generate labels and communicate them to upstream peers. * Communicate label-fec binding to peers once next-hop has been recognized * LSP is formed as incoming and outgoing labels are spliced together Comparison * Labels can be exchanged with less delay * Does not depend on availability of egress node * Granularity may not be consistent across the nodes at the start * May require separate loop detection/mitigation method Outgoing Label Ordered LSP control * Label-FEC binding is communicated to peers if: - LSR is the egress LSR to particular FEC - label binding has been received from downstream LSR * LSP formation flows from egress to ingress * Requires more delay before packets can be forwarded along the LSP * Depends on availability of egress node * Mechanism for consistent granularity and freedom from loops * Used for explicit routing and multicast 25

26 Ordered control must be used if an LSP shall have specific attributes, such as resources to guarantee QoS, specific route for traffic engineering considerations etc. In independent control parts of an LSP may be set up while others not. 26

27 Independent control + Downstream on Demand An LSR may answer requests for label mapping immediately, without waiting for a label mapping from the next hop. Independent control + Unsolicited Downstream An LSR may advertise a label mapping for a FEC to its neighbors whenever it is prepared to label switch that FEC. Ordered control + Downstream on demand An LSR assigns a label to a FEC only after it receives one from its next hop for the FEC, or it is an egress LSR. It advertises the label to its neighbors only upon request. Ordered control + Unsolicited Downstream An LSR assigns a label to a FEC only after it receives one from its next hop for the FEC, or it is an egress LSR. It advertises the label to its neighbors whenever it is prepared 27 to label switch that FEC.

28 LSP Route selection 1. Hop by hop routing each LSR independently choose the next hop for each FEC. Hop by hop routed LSP an LSP whose route is selected using hop by hop routing. 2. Explicit routing A single LSR, generally the LSP ingress or the LSP egress, specifies several ( or all ) of the LSRs in the LSP. If a single LSR specifies the entire LSP, the LSP is strictly explicitly routed. If a single LSR specifies only some of the LSP, the LSP is loosely explicitly routed. 28

29 Label Distribution Protocol - LDP (RFC 3036) This protocol is designed to support FECs that are address prefixes. i.e. a FEC is the set of packets that are routed according to a particular entry in the (ordinary ) Forwarding Table. A packet with a particular IP DA is classified into a FEC that matches its address prefix entry in the (ordinary) Forwarding Table. LDP uses the ordinary unicast routing tables to decide on the LSP route. LSP for a particular address prefix Unicast routing path 29

30 Label distribution ensures that adjacent routers have a common view of FEC < - - > label bindings Routing Table: Addr-prefix Next Hop /8 LSR 2 Routing Table: Addr-prefix Next Hop /8 LSR 3 IP packet Label Information Base Label-In FEC Label-Out xx /8 17 For /8 use label 17 Label Information Base Label-In FEC Label-Out /8 xx Step 3: LSR inserts label value into forwarding base Step 2: LSR communicates binding to adjacent LSR Step 1: LSR creates binding between FEC and label value Common understanding of which FEC the label is referring to! Label distribution can either piggybacking on top of an existing routing protocol, or a dedicated label distribution (LDP) can be created 30

31 Label Distribution can take place using one of two possible methods LSR1 LSR2 LSR1 LSR2 Label-FEC Binding Request for Binding LSR2 and LSR1 are said to have an LDP adjacency (LSR2 being the downstream LSR) LSR discovers a next hop for a particular FEC LSR generates a label for the FEC and communicates the binding to LSR1 LSR1 inserts the binding into its forwarding tables If LSR2 is the next hop for the FEC, LSR1 can use the label knowing that its meaning is understood Label-FEC Binding LSR1 recognizes LSR2 as its next-hop for an FEC A request is made to LSR2 for a binding between the FEC and a label If LSR2 recognizes the FEC and has a next hop for it, it creates a binding and replies to LSR1 Both LSRs then have a common understanding Both methods are supported, even in the same network at the same time. For any single adjacency, LDP negotiation must agree on a common method 31

32 LDP contains messages for : 1. Discovery : announces and maintains the presence of an LSR in a network. 2. Session : used to establish, maintain, and terminate sessions between LDP peers (LDP adjacency). 3. Advertisement: creates, changes and deletes local mappings for FECs. 4. Notification: signal error information. * The neighbor discovery phase runs over UDP. The other phases use TCP. * Messages are built from TLV options and thus can be extended by the addition of new option types. 32

33 LSR Neighbor Discovery A LSR announces itself by sending Hello messages on each of the subnets to which it is physically connected. The Hello messages are sent to the all routers multicast address and are destined to a special UDP port (646) where a router implementing LDP is listening for Hello packets. Hello, DA=All routers:646 link 33

34 After a LSR learns on a neighboring LSR, it opens a TCP connection to this neighbor, and it can send/receive binding information to/from the neighbor. It is also possible to open a connection with a non physically connected neighbor LSR. In this case its address must be known by other means, e.g. configuration. LDP uses TCP in order to guarantee reliable data transfer, in the correct order, and to guarantee that information from old connections will not be accepted. LDP session over TCP link 34

35 FECs in LDP are defined by address prefixes, and in particular by host addresses. A packet is classified into a FEC according to the longest match rule. The rules at an ingress LSR : IntraDomain routing: 1. If the DA is of a host to which a FEC is defined, the packet is routed according to this FEC. 2. Otherwise, if there is a FEC that the DA belongs to its address prefix, the packet is classified into this FEC. If several are possible, the one with the longest match is chosen. InterDomain routing: 3. If a packet needs to arrive at an egress router R whose address is included in a FEC, the packet is associated with that FEC. 4. If no FEC is found the packet is forwarded by ordinary routing, i.e. a label is not attached to the packet. 35

36 Example : A BGP router establishes an LSP from itself to another BGP router by Ordered control/downstream on demand control. Label request Label mapping BGP router BGP router A 1 R1 2 3 B The label request messages are routed according to ordinary shortest path routing. 36

37 The IGP in the AS maintains a host route for each BGP border router. Each interior router distributes its labels for these host routes to each of its IGP neighbors. Suppose A receives a packet P such that the next hop of DA(P) is B. Also, R1 attached label L1 to the IP address of B. Then, A attaches L1 to P and transmits the packet to R1. Thus, P is transmitted to B over the LSP from A to B. 37

38 Same example with Independent control+unsolicited Downstream Label mapping BGP router 1 2 BGP router BGP router 3 BGP router 38

39 A more complex example : AS1 BGP router BGP router A 1 R1 2 3 B C l1 l2 Address prefix X L1, l1 L1 is popped. l1 is swapped with l2 39

40 B and C distribute a label for each address prefix that they distribute via EBGP and IBGP. E.g. for X they distribute l1 and l2 respectively, B to A via IBGP and C to B via EBGP. As before, the IGP for AS 1 maintains a host route for each BGP border router. Each interior router distributes its labels for these host routes to each of its IGP neighbors. E.g. R1 attaches L1 to DA(B). Suppose A receives a packer P such that DA(P) belongs to address prefix X. A transmits P to B, and B then transmits the packet to C. Thus, A attaches (L1, l1) to P and transmits the packet to R1. P is transmitted to B over the LSP from A to B. At B, B pops L1, swaps l1 with l2 and transmits the packet to C. This way the packet can travel over an LSP between ASes until it reaches the destination AS. 40

41 RSVP-TE ( Traffic Engineering ) ( RFC 3209 ) RSVP-TE is an extension to RSVP that enables the establishemnt of explicitly routed LSPs, with possible resource reservations. Packets are classified into LSPs using various criteria at the ingress LSR. Recall that in RSVP packets are classified to RSVP flows using fields in the packets. In RSVP-TE a packet can be classified also, e.g. by the interface over which it arrives. An LSP that is generated by RSVP-TE is denoted LSP Tunnel. An LSP Tunnel does not necessarily represents a flow with reserved resources. RSVP-TE is also used to establish LSPs for TE purposes. 41

42 An LSP is established using the Downstream-on-demand label distribution and Ordered control. The ingress LSR initiates the LSP establishment. Request Step 1 Request Step n Main features: Step 2n Label Step n+1 Label 1. Establishment of LSPs with or without QoS requirements. 2. Dynamically rerouting of established LSP Tunnels. 3. Observation of the actual route traversed by an established LSP. 4. Identification and diagnose of LSP Tunnels. 5. Preemption of an established LSP Tunnels under administrative policy control ( setup & holding priorities ). 6. Capability to perform Downstream-on-demand label allocation, distribution and binding. 42

43 The PATH messages are used to request to bind a label to a FEC. The RESV messages contain a label. The label is the incoming label at the sender of the packet, and the outgoing label at the LSR that receives the packet. The receiving LSR continues the process by allocating a new label and forwarding the RESV to the next LSR in the path to the transmitting source. This way resources can be allocated to LSPs, associated with labels. The PATH messages can also contain an explicit route for the LSP. This is in contrast to RSVP that is based on conventional IP routing ( shortest paths ). 43

44 PATH ( label req. ) PATH ( label req. ) LSR A LSR B LSR C RESV ( label y ) RESV ( label x ) Data ( label y ) Data ( label x ) LSR A LSR B LSR C 44

45 RSVP-TE supports P2P and P2MP LSPs. If a source transmits a PATH message without an explicit route, the LSP that is established is either P2P or P2MP according to the DA respectively. Resources are reserved in this case according to the FF style, i.e. for every source resources are reserved separately. If the PATH message contains an explicit route, then LSPs are only P2P, between pairs of (source, destination). The reservation style is FF. 45

46 Example: two sources and two destinations S1 R1 S2 R2 Each source transmits two PATH messages with an explicit route. Four LSPs are established, to two destinations. The only possible resource sharing is : (S1 R1, S2 R1) and (S1 R2, S2 R2) because we actually have two sessions. However, RFC3209 enables resource sharing only if the source and destination addresses are the same, for rerouting. 46

47 An important feature for TE applications: Rerouting Rerouting occurs when there is a topological change or when a new route is available for a more efficient use of the network resources. In RSVP-TE a new LSP is first established, the traffic from the old LSP is rerouted to the new LSP, and then the old LSP is deleted ( make before break ). What happens if the new LSP needs resources of the old LSP? Because of Admission Control considerations, it can happen that the new LSP can not be established before the old LSP is deleted. A possible solution the old and new LSPs reserve resources by the SE style. When the new LSP is established, the ingress LSR, which is common to both the old and new LSPs, makes a reservation as a different source but with the same tunnel ID. The egress LSR creates a reservation of SE style, and so the new reservation shares 47 resources with the old LSP on common links.

48 Constraint based Routing-LDP: CR-LDP ( RFC 3212 ) What is constraint based routing? * In ordinary routing protocols, as RIP or OSPF, the path between a source and destination is set to be the shortest one according to some metric. * In constraint based routing a path is set according to other considerations, such as TE and QoS. Every link has a set of attributes, e.g. bandwidth and delay in relation to QoS. Every (source, destination ) pair has a set of constraints. These constraints are expressed by the same terms as the links attributes. E.g. a path between a given pair of ( source, destination ) needs a minimum amount of bandwidth ( constraint ). Constraint based routing looks for paths that fulfil the constraints, and among all the possible paths chooses the shortest one according to some metric. 48

49 * Possible constraints - Minimum bandwidth - Administrative constraints : traffic is not allowed in certain links, or must pass only through a set of links. - combinations of the above. * After the path is determined, it is necessary to set the route in the routers along the path. 49

50 * MPLS can be used to route packets along their chosen route because forwarding is according to a label and not according to the DA ( and so along shortest paths ). Also, the source knows which packets shall enter the LSP, in the same way that it knows which packets are to be routed according to Constrained based Routing. * The LSP is generated along a predefined path, which is not necessarily according to the shortest path ( as LDP works ). 50

51 CR-LDP is based on LDP with extensions that are mainly targeted to enable explicitly routed LSPs ( recall that LDP is based on unicast routing ). The extensions: 1. Explicit routing : the ability to create a LSP along a given path. Normally a list of IPv4 addresses ( source route ) that the Label Request message will follow. 2. Specification of the Traffic parameters of the traffic entering the LSP. 3. Resource Reservation and Resource Classes. 4. Route pinning & re-optimization. 5. Path preemption. 6. Handling failures. 7. LSP ID A unique tunnel identifier within an MPLS domain. Resource Class : resources are classified into several categories ( colors ). When an LSP is established, it is possible to define what colored resources it shall use. E.g. links have different colors. 51

52 * In CR-LDP the ingress LSR initiates the establishment of an LSP. * CR-LDP supports point-to-point LSPs only. * CR-LDP uses the Downstream-on-demand label advertisement with Ordered control. * In CR-LDP it is possible to define an Explicit Route ( ER ). This route is sent in Label Request messages. E.g. say LSR1 wants to establish the path LSR1, LSR2, LSR3, LSR4. It transmits a Label Request message to LSR2 with the path. Label Request Label Request Label Request LSR 1 LSR 2 LSR 3 LSR 4 Label Mapping Label Mapping Label Mapping 52

53 LSR2 finds that it is on the path, deletes itself, and sends the Label Request to the next LSR on the path, LSR3. LSR3 process the Label Request in a similar way. When LSR4 receives the Label Request it notices that it is the last LSR on the path. LSR4 allocates a label and transmits it to LSR3 in a Label Mapping message. LSR3 keeps the received label, allocates a new one and transmits its label to LSR2. This process continues up to the ingress LSR. 53

54 CR-LDP Traffic Parameters U F Traf. Param. TLV Length Flags Frequency Reserved Weight Peak Data Rate ( PDR ) Peak Burst Size ( PBS ) Committed Data Rate ( CDR ) Committed Burst Size ( CBS ) Excess Burst Size ( EBS ) 54

55 In CR-LDP it is possible to define the traffic parameters of the traffic flow that will be traversing the LSP. According to these parameters, resources are reserved for the LSP. CR-LDP follows after Frame Relay traffic parameters. CDR - Committed Data Rate ( bytes / sec ) CBS - Committed Burst Size ( bytes ) Define the rate that the MPLS domain commits to be available to the LSP. Both are parameters of a Token Bucket. CDR is the data rate which the user expects to pass into the network at all times and which the network should be able to handle. EBS - Excess Burst Size ( bytes ) Used to measure the extent by which the traffic sent on an LSP exceeds the committed burst. This is the size of a bucket in a Token Bucket. The token rate is CDR. 55

56 The CBS and EBS buckets are full at time 0. Let Tc and Te denote the number of bytes in the buckets respectively. Tc and Te are updated CDR times per second as follows : If Tc is less than CBS, Tc is incremented by one, else If Te is less than EBS, Te is incremented by one, else neither Tc or Te is incremented. CDR CDR Tc CBS Te EBS 56

57 When a packet of B bytes arrives for transmission : If Tc>B the packet is not in excess of the Committed rate and Tc is decremented by B. Else, If Te>B the packet is in excess of the Committed rate but is not in excess of the EBS and Te is decremented by B. Else, The packet is in excess of both the Committed rate and the EBS and neither Tc or Te is decremented. 57

58 PDR - Peak Data rate ( bytes /sec ) PBS - Peak Burst Size ( bytes ) Define a token bucket that controls the max. rate of traffic that can be transmitted into the LSP. PDR is the token rate and PBS is the bucket size. E.g. the PDR can stand for the physical rate of the line connecting the user to the network. Traffic above the PDR is subject to deletion by the network and there is no guarantee of delivery. It is useful for resource allocation. If the network uses the peak rate for resource allocation then its edge function should regulate the peak rate. 58

59 The PBS token bucket is initially full. Thereafter, the token count Tp, if less than PBS, is incremented by one PDR times per second. When a packet of size B bytes arrives at time t, the following happens : If Tp(t)-B >=0, the packet is not in excess of the peak rate and Tp is decremented by B down to the minimum value of 0 ; else The packet is in excess of the peak rate and Tp is not decremented. PDR Tp PBS 59

60 Frequency - the time period over which the CDR is guaranteed by the network to the LSP. Can receive three values : VeryFrequest over a short time period Frequent over a longer time period Unspecified over any time period. Weight determines the LSP s relative share of the possible excess bandwidth above its committed rate. The definition of relative share is MPLS domain specific. The traffic parameters are carried in the Label Request messages. They are returned in Label Mapping messages if they were negotiable. In this way LSRs can adjust the resources they reserved for the LSP in case the parameters were modified by downstream LSRs. 60

61 * Preemption If an LSP needs resources behind those that are free, it can take resources already allocated to other LSPs. Two priorities are defined : Set up priority Holding priority An LSP can take resources from an established LSP if its Set up priority is higher than the established LSP Holding priority. 61

62 Comparison between RSVP-TE and CR-LDP 1. Signaling RSVP is based on soft state * information is automatically cancelled in case of failures. * requires periodic update of the information. CR-LDP is based on hard state. 2. Reliability RSVP is based on periodic updates to achieve reliability. CR-LDP is based on TCP. So, no periodic updates are needed. However : * There is a traffic rate limit due to the congestion control mechanism of TCP. * There is a FIFO and thus an important message cannot be transmitted before less important ones. * The three-way handshaking delays LSPs establishment. This is important in case of a need for a fast-reroute. * LSRs that do not support TCP cannot use CR-LDP. 62

63 * The periodic overhead of RSVP can be significant on core networks ( a new proposal refresh reduction extensions is intended to reduce this overhead ). However, in CR-LDP one needs to find and handle all the possible failures that disable the use of a LSP in order to cancel the information of the LSP from the LSRs along the path. 63

64 3. QoS models RSVP is based on IntServ. CR-LDP is based on a new QoS model, as described above. IntServ was designed to be implemented over many L2 technologies. CR-LDP uses many parameters that are suitable to ATM and FR but with no clear mapping to other L2 technologies. 4. Security In CR-LDP messages are destined to the next hop and thus can be protected by IPSec or SSL. In RSVP PATH packets are destined to the egress LSR but are processed by intermediate LSRs. Therefore, IPSec cannot be used, and other security mechanisms shall be used. 64

65 MPLS & DiffServ ( RFC 3270 ) MPLS + DiffServ Domain The MPLS domain is composed of LSRs that only examine the label in their forwarding decisions. They do not consider the IP header at all! Thus, how the DSCP will be known for the correct treatment from QoS perspective? The DSCP is either 1. Written in a place where the LSR can find it, without exploring the IP Header, or 2. Shall be determined from the label. 65

66 1. When the label is carried in a Shim header then a three bit field, denoted Experimental, in this header, can be used. If the domain supports no more than 8 PHBs, these can be encoded in the Exp. Field. 2. If the domain supports more than 8 PHBs, the Exp. Field can not contain the DSCP and it needs to be determined somehow from the label. This is also the case when the label is encoded in a link layer header and not in a Shim header. First solution attempt : a label is attached to a (FEC,DSCP) pair. This means that packets of the same FEC but with different DSCPs are mapped to different LSPs. An ingress LSR decides on the LSP according to the FEC of a packet, and its DSCP (QoS requirements). 66

67 A problem with the first attempt : there is a requirement in the DiffServ standard which does not allow to misorder packets from the same Forwading Class which defer only in their drop priority. Ingress LSR AF11 AF12 Egress LSR AF13 67

68 Definition: PHB scheduling class : A group of PHBs that require that packets with different PHBs in the group should not be misordered, e.g. AF11, AF12, AF13. This means : packets that belong to the same FEC, with DSCPs from the same PHB scheduling set, must travel on the same LSP. Otherwise, misordering can happen because in principle different LSPs for the same FEC can be routed through different paths. Conclusion: the label itself cannot distinguish between PHBs in the same PHB scheduling set. 68

69 Therefore, one LSP for all packets in the same FEC and PHB scheduling class. How to differentiate among the different PHBs? If the Shim Header is used, by the Exp. Field, e.g., for the AF classes, AFxy, the x is encoded in the label and the priority class y in the Exp. field. Any existing LDP can be extended to specify which PHB ( or PHB scheduling set ) is to be bound to the advertised label, e.g. in the LDP protocol a label is now bounded to ( FEC, PHB ) pairs ( A FEC is an address prefix ). When an ingress LSR injects packets into an LSP it encodes the FEC and PHB scheduling class in the label, and encodes the dropping precedence in the Exp. bits. When there is no Shim Header : several labels are used for the same LSP, each encoding a different DSCP? 69