The document discusses the benefits of adding IP support to the Cello Packet Platform (CPP). As data traffic surpasses voice, networks will need to be packet-oriented while supporting delay-sensitive services like voice and video. While ATM was seen as the solution for quality of service, IP routing is now being introduced into telecom networks due to scalability, administration and cost issues with ATM. CPP is well-positioned to handle the new IP environment with features like DiffServ and MPLS support. The introduction of IP will start in core networks but eventually extend to access networks, so telecom vendors need solutions that operators can cost-effectively maintain during migration.
The document discusses the benefits of adding IP support to the Cello Packet Platform (CPP). As data traffic surpasses voice, networks will need to be packet-oriented while supporting delay-sensitive services like voice and video. While ATM was seen as the solution for quality of service, IP routing is now being introduced into telecom networks due to scalability, administration and cost issues with ATM. CPP is well-positioned to handle the new IP environment with features like DiffServ and MPLS support. The introduction of IP will start in core networks but eventually extend to access networks, so telecom vendors need solutions that operators can cost-effectively maintain during migration.
The document discusses the benefits of adding IP support to the Cello Packet Platform (CPP). As data traffic surpasses voice, networks will need to be packet-oriented while supporting delay-sensitive services like voice and video. While ATM was seen as the solution for quality of service, IP routing is now being introduced into telecom networks due to scalability, administration and cost issues with ATM. CPP is well-positioned to handle the new IP environment with features like DiffServ and MPLS support. The introduction of IP will start in core networks but eventually extend to access networks, so telecom vendors need solutions that operators can cost-effectively maintain during migration.
The document discusses the benefits of adding IP support to the Cello Packet Platform (CPP). As data traffic surpasses voice, networks will need to be packet-oriented while supporting delay-sensitive services like voice and video. While ATM was seen as the solution for quality of service, IP routing is now being introduced into telecom networks due to scalability, administration and cost issues with ATM. CPP is well-positioned to handle the new IP environment with features like DiffServ and MPLS support. The introduction of IP will start in core networks but eventually extend to access networks, so telecom vendors need solutions that operators can cost-effectively maintain during migration.
terized by change. Yesterdays dominant traffic typevoiceis being superceded by data traffic. In future telecommunications networks, voice will occupy only a small portion of bandwidth. This will put new de- mands on the networks, which will have to be packet-oriented and at the same time able to handle delay-sensitive traffic, such as voice and video (video conferencing). ATM technology has long been considered the so- lution to quality of service (QoS) in net- works, but several areas of concern have since been identified in large and growing net- works. These are scalabilitythe cost of the extra ATM layer makes it difficult for ATM vendors and operators to increase the link speed above 1 Gbit/s; administration and maintenancein large ATM networks, the number of per- manent virtual circuits (PVC) for router interconnections increases significantly, and thus the administration and mainte- nance of these PVCs becomes a major issue when networks need to be upgraded. A similar problem exists in pure IP net- works, but the addition of ATM does not simplify matters. In this sense, ATM is an extra layer that greatly increases com- plexity; and costfor example, the cost of having to support and maintain two kinds of net- work equipment. These issues have become a driving force for introducing IP routing into telecommuni- cations networks. The Internet boom has been accompanied by increased investments in IP technology, and a lot of effort has gone into solving quality-of-service and routing administra- tion issues. Multiprotocol label switching (MPLS), differential services (DiffServ, Box B), multiclass extension (MCE), and header compression have made it possible to replace ATM with pure IP networks. And CPP is well positioned to handle the new environ- ment (as well as the transition to it). DiffServ and MCE will be implemented in the first IP release of CPP, and the archi- tecture is ready to make use of MPLS. 68 Ericsson Review No. 2, 2002 CPPCello packet platform Lars-rjan Kling, ke Lindholm, Lars Marklund and Gunnar B. Nilsson C P P is a c a rrie r-c la ss te c h n o lo g y th a t h a s b e e n p o sitio n e d fo r a c c e ss a n d tra n sp o rt p ro d u c ts in m o b ile a n d fix e d n e two rk s. I t is a n e x e c u tio n a n d tra n sp o rt p la tfo rm with sp e c ifie d in te rfa c e s fo r a p p lic a tio n d e sig n . T h e e x e c u tio n p a rt c o n sists o f su p p o rt fo r th e d e sig n o f a p p lic a tio n h a rd wa re a n d so ftwa re . T h e tra n sp o rt p a rt, wh ic h c a n b e se e n a s a n in te rn a l a p p li- c a tio n o n th e e x e c u tio n p la tfo rm , c o n sists o f se ve ra l p ro to c o ls fo r c o m - m u n ic a tio n , sig n a lin g , a n d E T tra n sm issio n . T yp ic a l a p p lic a tio n s o n c u r- re n t ve rsio n s o f C P P in c lu d e th ird -g e n e ra tio n n o d e sR B S s, R N C s, m e d ia g a te wa ys, a n d p a c k e t-d a ta se rvic e n o d e s/h o m e a g e n ts. C P P wa s first d e ve lo p e d fo r a syn c h ro n o u s tra n sfe r m o d e A T M a n d T D M tra n sp o rt. N o w, su p p o rt is b e in g a d d e d fo r I P tra n sp o rt. T h e a u th o rs d e sc rib e th e te c h n ic a l a n d c u sto m e r b e n e fits o f a d d in g I P su p p o rt in C P P , wa lk in g th e re a d e r th ro u g h th e b a sic p rin c ip le s fo r I P se r- vic e s in C P P a n d th e C P P I P a rc h ite c tu re , wh ic h is ve ry ro b u st a n d sc a l- a b le . API Application program interface ATM Asynchronous transfer mode BGP Border gateway protocol CPP Cello packet platform DiffServ Differential services E1/J1/T1 PDH transmission frame formats for 2 Mbit/s (E1) or 1.5 Mbit/s (J1/T1) transmission rates ET Exchange terminal FIB Forwarding information base HA Home agent IP Internet protocol IPv4, IPv6 IP version 4, IP version 6 MCE Multiclass extension MGW Media gateway MPLS Multiprotocol label switching O&M Operation and maintenance OC3 Optical carrier 3 (155 Mbit/s) OSPF Open shortest path first PBA Printed board assembly PDSN Packet-data service node PPP Point-to-point protocol PVC Permanent virtual circuit QoS Quality of service RBS Radio base station RIP Routing information protocol RNC Radio network controller RSVP-TE Resource reservation protocol traffic engineering SCTP Stream control transmission protocol SIGTRAN Signaling transport SS7 Signaling system no. 7 STM-1 SDH transmission frame format for 155 Mbit/s TCP Transmission control protocol TDM Time-division multiplexing UDP User datagram protocol BOX A, TERMS AND ABBREVIATIONS Ericsson Review No. 2, 2002 69 Who are the customers? The introduction of IP in telecommunica- tions networks will commence in the core network in response to demands for large, high-speed networks. Datacom vendors are currently trying to grab as many market shares as they can. At present, these vendors can solely offer pure IP router products. Therefore, network operators can either opt to integrate telecommunications equip- ment and IP routers themselves or they can buy complete site solutions from a telecom- munications vendor. Over time, the IP mi- gration will extend further and further from the core network into the access network. Ericsson knows that operators want vendors to come up with solutions and migration sto- ries that are cost-effective and easy to main- tain (Figure 2). The introduction of IP in Applications own interfaces IP links ATM links TDM links IP links between Cello nodes and Ethernet PPP IP UDP TCP SCTP M3UA SCTP Applications Cello node Cello node Cello platform Switching Control system O&M Transport Operator interface Node hardware and software interfaces F ig u re 1 A n e two rk vie w o f th e C e llo p a c k e t p la t- fo rm C P P ) in c lu d in g tra n sp o rt p ro to c o ls. GSM/UMTS/GPRS core network BSC GSM-BSS UTRAN RNC PLMN PSTN/ ISDN Internet/ intranets MGW SGSN MGW GMSC server MSC server HLR Telecom applications IP ATM BB GGSN User traffic Signaling F ig u re 2 A te le c o m m u n ic a tio n s n e two rk with d iffe r- e n t typ e s o f b a c k b o n e a n d a c c e ss n e t- wo rk s. 70 Ericsson Review No. 2, 2002 Traditional routers serve packets on a first- come-first-served basis. Since no differentia- tion is made with respect to the actual demands put on the delivery times of packet flows, this is often called best-effort service. In a best-effort network, packet flows that carry delay-sensitive voice receive the same service as packet flows that carry, say, a file transfer. Different approaches have been proposed to solve this shortcoming. According to one approach, called the IntServe model, a flow must reserve resources in the routers on the path before packets can be sent. If sufficient network resources are not available, the flow is denied. An implication of IntServ is that every router in the network must store a per- flow state. This was seen as a major drawback and led to the introduction of a lightweight model called differentiated services (DiffServ). The DiffServ model is based on the following basic principles: A limited number of service classes is defined for specific purposes. These classes differ in terms of maximum delay or maximum drop probability. At the edge of a network, packets are marked in the header to reflect the service class to which the packet flow belongs. The routers in the network serve the packets according to its service class. In principle, packet flows that belong to the same service class are merged into a com- mon aggregated flow. The routers see these aggregated flows, but not individual packet flows. A packet enters a classifier, where the service class mark is inspected. Depending on the mark, the packet is sent to one of several queues via an enqueuer. Depending on the filling in the queue, the packet might be discarded by the enqueuer according to a buffer management algorithm. Using a scheduling algorithm, a scheduler fetches packets one-by-one from the queues. The buffer management and scheduling algorithms are configured to guarantee a specific per-hop-behavior for each service class, of which there are three standard classes: One best-effort class (BE) and two assured-forwarding classes (AF). Each assured-forwarding class is guaranteed a cer- tain minimum bandwidth at which the queues are served. Excess bandwidth is distributed to other classes. The best-effort class is served only if there are no packets in the other queues. If a queue is almost full, incoming packets to that queue are discarded in a pseudo-random fashion. In terms of resources, the DiffServ model causes the network to behave as if a few dif- ferent logical networks were separated from each other. For example, one logical network could be used for voice traffic and another for file transfers. BOX B, DIFFERENTIATED SERVICES E T-F M b o ard M P b oard M P b o ard E T-F M b o a rd S T M -1 S T M -1 E 1 /J 1 /T 1 E 1 /J 1 /T 1 M P b o a rd D P b o a rd S p a c e sw itc h M S B b o a rd M S B b o a rd FM F M F M F M E T -F M b o a rd FM F M F M Routing protocols IP security control IP security encryption IP host IP host Media stream device SIGTRAN IP host Gigabit Ethernet F ig u re 3 E m b e d d e d a n d d istrib u te d I P ro u te r. Ericsson Review No. 2, 2002 71 telecommunications networks adds yet an- other transport protocol. Today operators are struggling with the migration from TDM to ATM. Later, when operators introduce IP, they will have to consider even more complex networks. Ericsson can offer telecommunica- tions nodes with built-in transport capabili- ties for TDM, ATM and IP. Operators can thus capitalize on their installed base, espe- cially in mobile networks. Likewise, Ericsson knows that network operators want a compact and cost-effective solution that has a consis- tent network-management interface. Ericsson will thus make pure IP routers obsolete in op- erator networks. CPP is the ideal choice of car- rier-class technology for all telecommunica- tions nodes in the access network, including the edge nodes to the core backbone. Basic principles Six basic principles for the IP services in CPP add value: embedded and distributed IP router by means of routing protocols in the main processor cluster and distributed for- warding on all device boards (Figure 3); fully distributed forwarding of IPv4 or IPv6 modules can be implemented in hardware or software. A hardware imple- mentation can handle wire-speed trans- port (Figure 4); internal IP hosts can be located on a sin- gle printed board assembly (PBA) and connected to the IP router via the local forwarding module on the PBA (Figure 5). The IP host provides the application program interface (API); the internal hosts are IP end-systems that are visible and addressable from the ex- ternal network and from within the node (Figure 5); a CPP node can house multiple virtual routers. In this case, each IP interface in the node is configured to belong to one specific virtual router; and CPP has built-in support for signaling system no. 7 (SS7) signaling gateway functionality by means of a complete SS7 stack for SIGTRAN, IP, ATM and TDM (Figure 6). Architecture The CPP IP architecture is composed of sev- eral subsystems that interwork with each other through well-defined interfaces (Fig- ure 7). The IP forwarding subsystem pro- vides fully distributed forwarding of IPv4 GPB MP host FM FM FM FM Support for wire-speed user data by means of hardware support in the forwarding modules Host 1 Host n CPP node ET-board Application board UDP/IP termination support ET-board F ig u re 4 F u lly d istrib u te d I P fo rwa rd in g . SCCP NIF MTP3b SAAL AAL5 ATM Layer 1 MTP3 MTP2 TDM Layer 1 M3UA SCTP IP Layer 2 Layer 1 F ig u re 6 B u ilt-in su p p o rt fo r sig n a lin g g a te wa y. BP FM All hosts communicate via the local forwarding module Host 1 Host 2 Host 3 Host 4 Host x CPP node Application board All hosts have their own IP address and are visible in the IP network F ig u re 5 M u ltip le in te rn a l I P h o sts. and IPv6 packets. Forwarding modules (Box C) implemented in software or in ded- icated hardware circuits can be put on any board in the CPP system and are intercon- nected by the CPP space switch. Hardware- based forwarding modules can achieve wire- speed forwarding. The IP forwarding sub- system also includes packet classification and filtering, including DiffServ queuing. To forward an IP packet, the forwarding modules need a forwarding table that is cal- culated in the IP routing subsystem using static (configured) routes and dynamic rout- ing protocols, such as the open shortest path first (OSPF), routing information protocol (RIP), and border gateway protocol (BGP). The routing protocol handlers monitor the network topology, and the routing table manager calculates a forwarding table at start-up and when the network topology changes. Forwarding information from the IP routing subsystem (Box D) is communi- cated to all forwarding modules in the CPP node using the dedicated forwarding infor- mation base (FIB) interface. Telecommunications applications that use CPP IP transport services need to ter- minate and originate large amounts of IP traffic that comes from or is sent to the IP network. The IP access subsystem enables applications to use IP hosts in CPP. A user can access multiple IP hosts on every proces- sor board in the system: each host is identi- fied by a unique IP address. The hosts can handle IPv4 and IPv6, user datagram pro- tocol (UDP), transmission control protocol (TCP), and stream control transmission pro- tocol (SCTP) termination. If necessary, the hosts can be made robust by means of the moveable host concept (see also section on Robustness). An automatic configuration mechanism simplifies host configuration. The IP access subsystem uses the IP forwarding subsystem to send and receive packets. For low-speed applications, CPP provides a distributed host mechanism whereby one IP host with one IP address can be distributed and used on multiple processor boards. IP transport is also used for highly confi- dential traffic, such as control signaling and operation and maintenance (O&M) traffic. To guarantee the integrity of this traffic, the IP security subsystem provides tunnel- and transport-mode encryption and decryption of IP packets. The IP security encryp- tion/decryption engines can be distributed on multiple processors or dedicated hard- ware circuits to yield greater capacity. The CPP architecture has been prepared for the introduction of MPLSCPP can serve as a label edge router or label switch router. However, modifications will need to be made to the forwarding modules, and ad- 72 Ericsson Review No. 2, 2002 Forwarding modules implement the function- ality needed to forward or terminate an IP packet. They also provide the functionality for resolving dependencies on connected routers (the data part) and hosts (the control part). Figure 8 shows the data part of a wire- speed forwarding module. Most of the func- tionality is implemented in dedicated hard- ware or network processors. In addition, an exception handler executes on a standard processor. The exception handler provides the functionality needed for dealing with some infrequent packet types. If, during any step in the forwarding process, an exception is identi- fied, the packet is handed over to the excep- tion handler. Packets enter the forwarding module from an incoming link. Each packet is classified according to the type of service it requires. This classification might involve metering actual load. The type-of-service field in the packet might be changed as a result of the classification. The classification also serves as a sort of firewall, and might result in having the packet dropped. The packet is next analyzed to find out if it should be terminated in the node. If so, a point of termination in the node is determined and the packet is put into one of several queues to the switch. Otherwise, the packet is further analyzed to determine the best next hop in the network; and various related parameters. One parameter contains the outgoing link and the point in the node where the forwarding module responsible for this link is located. The packet is then put into one of the switch queues. After having passed the switch, the packet arrives at the forwarding module of the appointed outgoing link where it enters a second classification step, which might result in the packet being assigned a different class. Finally, the packet enters the queuing sys- tem to the outgoing link. The queuing system has several queues served by a configurable, weighted, fair-queuing scheduler. The system also has advanced dropping mechanisms to handle overload situations on the link. The treatment a packet receives in the queuing system is determined by the class to which the packet has been assigned. BOX C, FORWARDING MODULES OAM OAM OAM OAM OAM OAM OAM OAM IP routing IP forwarding IP access IP applications ATM applications IP security M PLS IP link layer Physical link ATM transport services CPP platform F ig u re 7 T h e C P P I P a rc h ite c tu re . Ericsson Review No. 2, 2002 73 ditional signaling protocols will be intro- duced, including the resource reservation protocol traffic en- gineering (RSVP-TE); and CR-LDP. Likewise, interworking will be introduced between IP routing and MPLS. Together with the physical layer sub- system, the IP link layer subsystem provides access to the external network. For example, the IP link layer subsystem supports 10/100 Mbit Ethernet, Gigabit Ethernet, point-to-point protocol (PPP), IP over ATM, and frame relay. Likewise, the phys- ical layer supports STM-1/OC3, E1/J1/T1 and Ethernet. The IP link layer functional- ity is connected to the IP forwarding sub- system using the generic link interface. Scalability CPP is uniquely scalable. It can be used in small applications (such as a small RBS) as well as large RNC and media gateway nodes. Indeed, virtually every aspect of scalability is covered by CPP: high-end and low-end nodes, payload capacity, processing capaci- ty, number of routes, route updating capac- ity, number of physical links, link capacity, and cost. In general, the IP functionality of CPP is composed of central functions and local functions. A central function solely exists in one instance for each virtual IP router, whereas a local function might exist in sev- eral instances. With few exceptions, the scaling of local functions is quite straight- forward. By contrast, the scaling of a central function is an intricate matter since central functions have a single, consistent informa- tion base. Distribution is not precluded but must be accomplished without incurring excessive costs in terms of memory and ca- pacity. P a ylo a d h a n d lin g c a p a c ity CPP forwarding is based on interacting for- warding modules. Each forwarding module is a local function that handles a number of links and interfaces. Each processing entity (processor) and printed board assembly has its own forwarding and link module, which means that forwarding and link capacity are not adversely affected when new hardware is added. The only potential bottleneck is the intercommunication capacity of the for- warding module, which is based on the CPP space switch. At present, the capacity with- in the subrack is 16 Gbit/s, which is suffi- cient to support some 400,000 IP-borne voice calls. Moreover, multiple subracks can be interconnected to form very large nodes that constitute a single coherent system. As with forwarding, IP access mainly consists of local functions and scales smoothly when new hardware is added. F o rwa rd in g in fo rm a tio n b a se The number of routes to be handled direct- ly affects the size of the forwarding infor- Classification m etering Forwarding m odule Classification m etering Classification m etering Term ination analysis Term ination analysis Forwarding Forwarding Switch queue Exception handler Switch Link queue Link F ig u re 8 I n te rn a l a rc h ite c tu re o f th e fo rwa rd in g m o d u le . mation base (FIB). A very large FIB makes it unfeasible to house the entire forwarding table in fast path-forwarding logic. There- fore, CPP provides a caching scheme which forwards packets that fall outside the scope of the cache. At present, the fast path- forwarding path supports up to 64,000 routes. R o u tin g p ro to c o ls The central functions that demand the most processing power in CPP are the routing protocol handlers (OSPF, BGP and so on). A shortage of processing power for routing results in unacceptably long convergence timethat is, the time it takes to regener- ate a routing table when the topology has changed. Certain measures affecting network layer configuration can significantly reduce the processing load on the individual routers. These measures are hierarchical network topology, appropriate address allocation, route summarization, and area subdivision. CPP supports these measures as well as scal- ability at the node level. A llo c a tio n o f ro u tin g p ro to c o l h a n d le rs Different routing protocol handlers (RPH) can be allocated to different processors. Every RPH is mastered by a single routing table manager. However, other scaling so- lutions are necessary when a single RPH re- quires more capacity than can be provided by one processor. A d d in g a virtu a l ro u te r Where the network is concerned, the addi- tion of a virtual router is no different from the addition of a physical router. In either case, the network becomes more complex and the RPHs must work harder to keep track of the network topology. The advan- tages in the local node are that the interfaces can be partitioned between virtual routers, and functions that belong to different vir- tual routers can be allocated to different processors. In addition, costs can be avoid- ed provided that the virtual-router inter- connecting link is implemented efficiently. D istrib u tin g a sin g le R P H A single RPH can be distributed to some extent by allocating certain sub-functions to different processors. The exact allocation of sub-functions varies according to the types of routing protocol in use. For OSPF, the shortest path calculation must be executed in one processor, whereas the handling of in- terfaces can be distributed. Low-end scalability For CPP to meet the requirements of radio base stations it must support low-end scal- ability. As a consequence, the CPP IP ar- chitecture has been designed to accommo- date IP functionality on a single PBA and to eliminate the switch. Another feature is the FIB caching mechanism mentioned above. These features keep the cost of fast forwarding-path logic to a minimum even in relatively complex networks with multi- ple routes. Robustness The fully distributed IP-forwarding and IP- termination mechanisms in CPP are very ro- bust. The failure of a board solely affects the forwarding and termination of IP packets on that board. All other boards continue for- warding and terminating as if nothing had happened, except that packets are not for- warded to the failed board. Should a net- work interface board or link fail, the rout- ing protocols initiate link protection switching or automatic rerouting. The robustness principles applied in CPP make it possible to define reliable programs that execute in the main processor cluster. These programs can be run on a standby processor should the main processor fail. The CPP IP functionality employs this fail-over concept for all central functions, such as the routing protocols, the central parts of IP ac- 74 Ericsson Review No. 2, 2002 The Internet is made up of several autonomous systems (AS), each of which is operated by an Internet service provider (ISP) or an organization (Figure 9). In telecommunications, each IP-based radio access network or core network can be seen as an autonomous system. Interior gateway routing protocols, such as RIP, OSPF or IS- IS, are used inside autonomous systems to automatically create forwarding tables to be used by the routers. OSPF is one of the most popular interior gateway protocols. This link state routing pro- tocol discovers its neighboring routers and learns their network addresses by sending out Hello packets on all its interfaces. The infor- mation it receives is stored in a topology data- base. OSPF then sends out a link state adver- tisement (LSA) packet indicating the status of its own links. The LSAs are flooded by the other routers in order to reach every router in the AS. Based on this input, OSPF calculates the shortest path to all routers and builds an optimal forwarding table. Whenever a change in network topology takes placethat is, if a link or a router goes downnew LSAs are flooded through the autonomous system. OSPF in each router updates the topology database and recalculates the forwarding table. Some autonomous systems can be very large and difficult to manage. Likewise, the LSA flooding can generate a heavy load in a large autonomous system. To reduce load and simplify management, OSPF allows autonomous systems to be divided into OSPF areas. Routers connected to one area need only have detailed knowledge about the topol- ogy of that area. All areas in an autonomous system are interconnected by the backbone area. The routers connected to more than one area are called area border routers. BOX D, IP ROUTING Ericsson Review No. 2, 2002 75 cess, IP security, IP forwarding, and all O&M functions. Should the processor on which the routing protocols execute fail, then the routing protocols are simply trans- ferred to the standby processor. In the in- terim, while the protocols are being trans- ferred and a new forwarding table is being built, the forwarding of packets on all other boards continues uninterrupted using the most recent forwarding table. In coming releases of CPP, the standby routing protocol will operate in a listening mode and maintain an updated standby routing table. The internal hosts in CPP, which appli- cations use for terminating and originat- ing IP traffic, can be distributed on any processor board in a CPP node. Each host is identified by a unique IP address. Should a board fail, CPP automatically transfers the host (and IP address) to another board. At the same time, the forwarding table is recalculated and traffic destined for the host is forwarded to the new location. If an application requires it to do so, CPP can configure a robust Ethernet IP interface that uses a primary and secondary physical port. Should the primary port fail, the sec- ondary port is activated using the IP address that the primary port had. Conclusion Ericsson knows that operators want vendors to offer solutions and migration stories that are cost-effective and easy to maintain. Con- sequently, Ericsson offers telecommunica- tions nodes with built-in transport capabil- ities for TDM, ATM and now, IP. Ericsson also knows that network operators want a compact and cost-effective solution that has a consistent network-management inter- face. Accordingly, Ericsson will make pure IP routers obsolete in operator networks. The CPP IP architecture is composed of several subsystems that interwork with each other through well-defined interfaces: The IP forwarding subsystem provides fully distributed forwarding of IPv4 and IPv6 packets. The IP routing subsystem calculates a for- warding table, which forwarding modules use to forward IP packets. The IP access subsystem enables applica- tions to use IP hosts in CPP. An automatic configuration mechanism simplifies host configuration. The IP security subsystem guarantees the integrity of sensitive traffic. The CPP architecture has been prepared for the introduction of MPLSCPP can serve as a label edge router or label switch router. Interworking will be introduced between IP routing and MPLS. Together with the phys- ical layer subsystem, the IP link layer sub- system provides access to the external net- work. CPP is uniquely scalable. It can be used in small applications and large RNC and media gateway nodes. Virtually every aspect of scalability is covered by CPP: high-end and low-end nodes, pay-load capacity, pro- cessing capacity, number of routes, route updating capacity, number of physical links, link capacity, and cost. The fully distributed IP-forwarding and IP-termination mechanisms in CPP are very robust. The robustness principles applied in CPP make it possible to define reliable pro- grams that execute in the main processor cluster. Backbone area Backbone area Area 1 AS R R R R R R R R R R R R AS BGP Area 2 F ig u re 9 E x a m p le o f se c tio n o f th e I n te rn e t c o n sistin g o f two a u to n o m o u s syste m s A S ) , wh e re o n e A S is d ivid e d in to th re e O S P F a re a s. T h e b o rd e r g a te wa y p ro to c o l B G P ) is u se d to c o n ve y ro u tin g in fo rm a tio n b e twe e n th e two a u to n o m o u s syste m s.