Service Providers Networks & Benefits of Multi Protocol Label Switching (MPLS) 20/11/2009 Local Team
Service Provider Networks & Carrier Networks A telephone company (or telco) provides telecommunication services such as telephony and data communications. Telcos are also known as common carriers. A service provider is an entity that provides services to consumers. Most telcos now also function as internet service provider (ISPs), and the distinction between telco and ISP may disappear completely over time, as the current trend for supplier convergence in the industry continues. Primary Line telephony POTS and other telephony related service Leased Lines Point-to-Point service Internet WEB1.1 based(today), WEB2.0 (very near future) VPN Customer Virtual Private Networks Today most of the services are delivered in different networks much rely on different physical infrastructure. WHAT IS CONVERGENCE?
Service Provider Networks TDM Networks TDM Time Division Multiplexing Circuit switch structure Each circuit is assigned to a timeslot in time domain A circuit should be pre-configured and remain provisioned for connectivity even when there is no data flow Waste of resources A popular example of TDM is GSM base station bachauling. 100% resources should be dedicated for a 5% overall usage (Todays GSM) Possible over-subscribing(10 channels for 20 users) may lead out of Service time Very reliable and secure since no users share same medium High cost per BW In the early 1990s, most networks were private line (or point-to-point), meaning a physical circuit ( combination of timeslots) had to be provisioned between locations. If a headquarters location needed to communicate with 10 other locations, the location needed 10 separate private lines.
Service Provider Networks ATM & FR ATM is a packet oriented transfer method that uses asynchronous (TDM) technique. No need to assign a dedicated physical channel to each information flow Virtual channels are introduced 53-Byte fixed length cell switching Sharing of physical resources between multiples of virtual channels (Overbooking) QoS is introduced via CAC Up to now there was no need for any prioritization and policing of traffic since no sharing exist With frame relay and ATM, the big difference was a logical connection for direct communications between locations. This greatly reduced costs. When organizations switched from private line to frame relay or ATM networks, the primary driver was reducing transport cost sometimes by more than 50%.
Service Provider Networks ATM & FR
What happened to Internet? The world's largest network of computer networks got its original name from the U.S. military arm that funded it: Arpanet was for the Advanced Research Projects Agency. Routers are the building blocks of Internet using IP at the control plane A hop by hop based architecture Routing protocols to discover paths When networks get too large, discovering and selecting routes becomes slow, ineffective. For this reason, by the late 1990s, most large ISPs had created twotier architectures, with an outer ring of intelligent routers communicating across a switched (typically ATM) core, implementing traffic engineering at the core.
IP - over - everything This approach worked well until new customer needs come on surface More types of services with more BW Immidiate access to any service, anytime and anywhere There are three problems with this approach to newly surfaced consumer needs First is the well known "cell tax," the bandwidth overhead resulting from segmenting large IP packets into 53-byte ATM cells. Increasing utilization of links with non-profit BW (increased CAPEX) In addition, service providers must manage and administer multiple networks of devices (optical transmission, ATM switching and IP routing), Increased OPEX. Inadaquate service differentiation capabilities ATM has only 3 types of flow classifier extended to 5 types of services in which IP has 64 Classes of service for end customer Current core networks almost use 8 classes of service With upcoming LTE 9 CoS will be required More the CoS is more consumer flexibility (Service Provider Perspective)
Raise of MPLS - 1 1. MPLS embraced IP In the early 1990s, the telecom industry was pinning all of its hopes on ATM as the network backbone technology of the future. But in 1995, the Internet exploded, and carriers had to quickly refocus their efforts in a different direction. By 1996, IETF researchers were looking for ways to make circuit-oriented ATM technology run over IP. ATM proponents jumped aboard the MPLS bandwagon in 1997, when the IETF formed its MPLS Working Group and MPLS team was wise to embrace rather than fight IP.
Raise of MPLS -2 2. MPLS is protocol neutral MPLS was designed to work in a multiple protocol environment. That allowed MPLS to work with ATM, Frame Relay, Sonet or Ethernet at the core.. MPLS also played a key role in supporting both legacy network technologies and the latest IP-based technology. Today, MPLS is being used to support metro-ethernet services, mobile communications back-haul communications and video distribution.
Raise of MPLS - 3 3. MPLS scales Successful Internet technologies need to be able to scale quickly, and MPLS was able to do that. Verizon uses MPLS for several global networks including its public and private IP networks Verizon s Public IP network, for example, spans 410 points of presence on six continents and spans more than 150 countries. These massive networks showed that "MPLS did work, and that it worked on a significant scale."
Service Provider Networks - MPLS MPLS = Multiprotocol Label Switching Objectives of MPLS Working Group: Enhance performance and scalability of IP routing Facilitate explicit routing and traffic engineering Separate control (routing) from the forwarding mechanism so each can be modified independently Develop a single forwarding algorithm to support a wide range of routing functionality
Example : Forwarding in IP network LPM lookup IP addr. Forward to next hop LPM lookup IP addr. Forward to next hop Access LPM lookup IP addr. Forward to customer A IP router Edge C B IP router IP router Core D IP router E IP router IP network view: A B C D E
Example : Forwarding in MPLS network LPM lookup IP addr. Insert label Forward to next LSR Exact match lookup Swap label Forward Access to next LSR Remove label LPM lookup IP addr. Forward to customer A Label Edge Router (LER) Label Switch Routers (LSRs) E Label Edge Router (LER) IP network view: A MPLS network E
MPLS Basic Components What is a label? a label is a short, fixed length, locally significant identifier that is carried by the packet and used to identify a FEC the generic solution for assigning a label to a packet is by insertion of the label between the network layer (IP packet) and the data link layer. This may look as follows OSI layering model 4 3 2½ payload payload payload IP Header IP Header MPLS header MPLS header MUST include : label or label stack MPLS header MAY include : TTL value stack indicator class of service 2 payload IP Header MPLS header Layer 2 header
MPLS Basic Components - Labels What does a label look like? This depends on L2/L1 protocol used For PPP data links and LAN data link (e.g. Ethernet): 32-bits IP Payload IP Header MPLS Header L2 Header Label (20-bits) EXP S TTL TTL value enables the following like in IP to Avoid loops: TTL = 0 drops the packet Limit the forwarding scope of the packet To reflect total number of hops, TTL value from IP packets is copied inside label and decreased at each LSR it passes through. At egress, LSP TTL may be copied back into IP TTL
MPLS Terminology Ingress LER/LSR Transit LSR Egress LER/LSR Label PUSH Label SWAP Label POP LSP: Label Switched Path Upstream Downstream
MPLS Forwarding Example Ingress Routing Table Destination Next Hop 134.5/16 LSP3 200.3.2/24 LSP5 MPLS Table In Out (2, 84) (6, 3) POP 134.5.6.1 PUSH 2 6 SWAP 134.5.1.5 2 134.5.1.5 LSP3 MPLS Table Destination Next Hop 3 LSP5 1 2 3 5 Egress Routing Table Destination Next Hop 134.5/16 134.5.6.1 200.3.2/24 200.3.2.1 LSP3 (2, 84) LSP5 (3, 99) MPLS Table In Out (1, 99) (2, 56) MPLS Table In Out (3, 56) (5, 3) 200.3.2.1 200.3.2.7
MPLS Label Distribution Protocols MPLS requires a signaling protocol to: Coordinate label distribution Explicitly route the LSP Bandwidth reservation (optional) Class of Service (DiffServ style) Resource re-assignment Pre-emption of existing LSPs Loop prevention MPLS signaling protocols defined by IETF Label Distribution Protocol (LDP) Resource Reservation Protocol (RSVP)
Label Distribution Protocol - LDP Upstream LDP peer Net: 10.0.0.0 Label: 17 LSR Net: 10.0.0.0 Label: 52 Downstream LDP peer Net: 10.0.0.0 Label: 29 3 1 4 5 2 3 MPLS Table Advertise MPLS Table In Out incoming In Out (1, 17) label (4, 17) (5, 52) IP Route Receive MPLS Table outgoing In Out label (2, 52) (3, 29) 10.0.0.0 Distributes label binding information Runs on LSRs in conjunction with IP routing protocols Labels are periodically refreshed Labels assigned by downstream peer Limitations LSPs follow conventional IGP path Does not support explicit routing
Resource Reservation Protocol - RSVP RSVP was originally designed for use in IP networks Enables end-to-end QoS reservation of resources for individual data flows (IntServ)- Integrated Services Requires all routers to maintain state of each micro-flow from source to destination Scalability issues limited deployment of RSVP to a few private networks Signaling component is now used for other applications Differentiated Services (DiffServ) MPLS Traffic Engineering
Resource Reservation Protocol - RSVP Ingress LSR PATH Explicit route = {R1, R4, R8, R9} Egress LSR R1 R4 R8 R9 RESV RSVP already has the resource reservation component built-in Makes it ideal to reserve resources for LSPs RSVP is structured, extensible protocol (TLV: Time, Length, Value) Proposed extensions are backward compatible with traditional RSVP implementations
Comparing Label Distribution Protocols LDP Hard state (TCP) Slow failure detection (IGP) Shortest path only No QoS or BW reservation Automatic LSP setup RSVP-TE Soft state - needs refresh Fast failure detection (hello timeout) Allows control of path Allows QoS, BW reservation Manual LSP setup (N-squared)
MPLS Path Protection Fast Re-Route Primary/Backup MPLS tunnel consists of Primary LSP and Secondary LSP (optional) Backup path calculation by constraint based routing algorithm or external tool No other routers in common with primary LSP Backup LSP can be cold standby or hot standby No traffic on backup LSP as long as primary LSP is up Primary LSP ingress egress Backup LSP
Traffic Engineering - CSPF Path calculation by constraint based routing algorithm or external tool Constraint Based Routing Algorithm (CSPF) TE extensions to OSPF and IS-IS Traffic engineering database created through OSPF or IS-IS extensions Used to exchange available bandwidth and color of links Constraints Bandwidth Administrative color Include-color-group: All the links that are chosen must have at least one color found in the include color group Exclude-color-group: All the links that are chosen must not have a color listed in the exclude color group Max number of hops Include strict/loose hops Avoid node(s): e.g. secondary LSP should avoid nodes/interfaces used for primary
MPLS DiffServ The DiffServ model offers a scalable solution for IP QoS in backbones MPLS has been enhanced in support of DiffServ: an IP packet s DSCP can be mapped in the MPLS header information (EXPbits, or EXP-bits and label) E-LSP L-LSP Separate LSP for each QoS class => support >8 QoS classes EXP field encodes drop precedence Finer granularity for TE and LSP protection Up to 8 QoS classes in a single LSP EXP field encodes DSCP Advantages compared to L-LSPs label space conservation less signalling overhead less consumption of forwarding state in LSRs TCP/IP host IP phone Server Access edge Core edge Access
DiffServ aware MPLS DiffServ CE DiffServ-aware MPLS TE in a DiffServ network PE MPLS PE LSRs advertise multiple available bandwidths via IGP P P Aggregate admission control against a particular bandwidth pool PE P P PE Packets should be routed based on expected QoS CE DiffServ
Why we need MPLS? ASIC design has improved tremendously in last decade Today, 10Gbps IP forwarding can easily be done in hardware Memory has become drastically cheaper in last decade No problem storing 1 Million destinations Today, MPLS does not offer faster forwarding than IP
Do we need QoS? Actually, most backbone IP networks today do not use ATM or MPLS for QoS Enforcing QoS is only important when there is congestion Perfect QoS can be provided in IP, if there is no congestion Just overprovision the network! Most IP networks keep bandwidth utilization at ~30-40% Is this cheaper or more expensive? Depends on the operator
Real reasons to deploy MPLS? Traffic Engineering: Manage traffic load on different parts of the network Differeciate Service approach Virtual Private Networks Offer point-to-multipoint services - IP VPNs and VPLS Replace traditional (point-to-point) FR/ATM services - VLLs New services such as VoIP or Video Introduction of New Service Tripple Play Service (Voice, Video and Data together) Differentiation between consumers within same service type
Rushmore Evolution Phase 2 HW & SW
Q&A www.alcatel-lucent.com 31 TiMOS-7.0R3 P2MP LSP workshop September 2009
www.alcatel-lucent.com www.alcatel-lucent.com 32 TiMOS-7.0R3 P2MP LSP workshop September 2009