ericsson

IntroductionIn the next few years, networks will evolvetoward todays vision of an all-IP network.We can see two possible scenarios: one fortraditional networks, and one for next-generation networks. For traditional net-works, the scenario describes what will hap-pen in multiservice networks and in second-

generation mobile networks, such as GSM.The next-generation network scenario de-scribes the development of mobile Internetand fast Internet access in fixed networks.

In traditional networks, evolution is dri-ven by never-ending growth in the need forprocessing capacity; in mobile networks, bygrowth in the number of subscribers andusage levels per subscriber. The wirelinenetwork is also experiencing a sharp increasein traffic because of Internet surfing.

In next-generation networks, traditionaltelephone and data networks will convergeto become general networks designed formany different services and modes of access.The convergence of data and telecommuni-cations makes it possible to combine the bestof two worlds. Some requirements, such asthe need for heightened performance, arefulfilled more easily when development isbased on data communications products.Also, the variety of access modesvia second- and third-generation mobile net-works, the multiservice networks, andbroadbandwill necessitate the coexis-tence of different transmission formats.Thus, gateways will be required at networkinterconnection points.1

10 Ericsson Review No. 1, 2001

Thanks primarily to an architecture that was developed to supportchange, the AXE exchange continues to evolve. Its architecture and mod-ularity have benefited customersAXE has served as local exchangesand international exchanges, and even in mobile networks to providemobile switching centers (MSC), home location registers (HLR), and otherfunctions. This has resulted in a total number of about 13,000 exchangesand an all-time-high growth rate.

The modularity of AXE makes it possible to add new functionality in acost-effective way, but hardware and software R&D must also make themost of new technologies.

This article describes recent adaptations of hardware and software thatwill prepare AXE for the next generation of networks. The authors focuson a system architecture that will serve as the basis for migration towarda server-gateway architecture for third-generation mobile networks andthe next generation of multiservice networks. Adaptations will also enableimprovements in existing networks where traffic is growing quickly.

AAL ATM adaptation layerACS Adjunct computer subsystemALI ATM link interfaceAM Application moduleAP Adjunct processorAPC AM protocol carrierAPIO AXE central processor IOAPSI Application program service

interfaceASIC Application-specific integrated

circuitATM Asynchronous transfer modeBICC Bearer-independent call controlBIST Built-in self-testBSC Base station controllerC7 CCITT (now ITU-T) no. 7, a com-

mon-channel signaling systemCAS Channel-associated signalingCP Central processorcPCI Compact peripheral component

interconnectCPP Cello packet platformDL Digital linkDLEB Digital link multiplexer board in the

GEMDSA Dynamic size alterationDTI Data tranmission interworkingECP Echo canceller in poolET Exchange terminalETSI European Telecommunications

Standards InstituteFTP File transfer protocol

GCP Gateway control protocolGDDM Generic datacom device magazine

(subrack)GDM Generic device magazine (subrack)GEM Generic Ericsson magazine (subrack)GS Group switchGSM Global system for mobile

communicationGSS Group switch subsystemHDLC High-level data-link controlHSB Hot standbyIO Input-outputIOG Input-output groupIP Internet protocolIPN Interplatform networkISDN Integrated services digital networkISP Internet service providerIWU Interworking unitMGW Media gatewayMIPS Million instructions per secondMSC Mobile switching centerMSCS Microsoft cluster serverMSP Multiplex section protectionMTP Message transfer partMUP Multiple position (timeslot)NGS Next-generation switchOSS Operations support systemPCM Pulse code modulationPDH Plesiochronous digital hierarchyPLMN Public land mobile networkPSTN Public switched telephone networkPVC Permanent virtual circuit

RAID Redundant array of independentdisks

RAM Random access memoryRISC Reduced instruction set computerRLSES Resource layer service specificationRM Resource moduleRMP Resource module platformRP Regional processorRPC Remote procedure callRPP Regional processor with PCI interfaceSCB-RP Support and connection board - RPSDH Synchronous digital hierarchySES Service specificationSONET Synchronous optical networkSS7 Signaling system no. 7STM Synchronous transfer modeSTOC Signaling terminal open communi-

cationTCP Transmission control protocolTDMA Time-division multiple accessTRA TranscoderTRC Transceiver controllerTSP The server platformTU 11, 12 Typical urban 11 (12) km/hrUMTS Universal mobile telecommunica-

tions systemVCI Virtual channel identifierVPI Virtual path identifierXDB Switch distributed boardXSS Existing source systemWCDMA Wideband code-division multiple

access

BOX A, TERMS AND ABBREVIATIONS

AXE 810The evolution continuesMagnus Enderin, David LeCorney, Mikael Lindberg and Tomas Lundqvist

Ericsson Review No. 1, 2001 11

Typical of next-generation network ar-chitecture is the separation of connectivityand communication or control services. Thisnew new architecture will appear first inGSM and WCDMA/UMTS core networks,where AXE exchanges will continue to serveas MSCs. For multiservice networks, tele-phony servers will become hybrid nodeswhich consist of an AXE exchange that usesan AXD 301 to handle packet-switcheddata.

Based on the traditional and next-generation scenarios, network evolutionwill demand increased processing capacity,greater switching capacity, and conversionto packet-switched technology. In addition,new ways of doing business will emerge asvirtual operators pay for the right to use tele-com equipment owned by other operators.Similarly, more operators are expected toenter the market, putting additional de-mands on charging and accounting func-tions.

To succeed in todays telecommunica-tions market, operators must be able to pro-vide their users with new functionality innetworks and complete coverage by mobilesystems at the same or a lower cost as timeprogresses. Operators put demands on thereturn on investment, cost of ownership,footprint (multiple nodes per site), plug-and-play functionality (short lead times inthe field), and quality of software packages(quality of service).

This article describes the latest develop-ments made in the AXE platform to meetthese requirements, and what new productswill be launched. For example, to shortenthe software development cycle, a layered ar-chitecture was introduced in the early1990s, resulting in application modules(AM). Current work on application modulesfocuses on the server-gateway split. AXEhardware is also undergoing a major archi-tectural transformation to further reduce thefootprint, power consumption, and numberof board types.

Ericssons goal is to cut the time to cus-tomer for AXE by targeting improvementsin production, transportation, and installa-tion. Far-reaching rationalization has beenachieved through the introduction of thegeneric Ericsson magazine (GEM), an open,flexible, scalable, high-capacity magazine(subrack). A new distributed, non-blockinggroup switch (GS890) has also been intro-duced, as have new devices, such as the ATMlink interface (ALI) and ET155-1, which en-able AXE to serve as a gateway to an ATM

network. Also, a new input-output (IO) sys-tem, called the APG40, has been developedusing products from mainstream data com-munications vendors.

The application platformThe traffic-handling part of AXE has a two-layer architecture: the application platform,which consists of hardware and software; andthe traffic applications, which solely consistof software. The application platform can beseen in terms of hardware and software,which present different but complementaryviews of the architecture.

The software viewThe AXE application software has contin-ued to evolve since it was first layered andmodularized in the early 1990s. Market de-mand for shorter lead-times was the mainforce driving the change to layers and mod-ularization.

Application modules

Traffic applications are encapsulated in ap-plication modules that use the application-platform softwarecalled the resourcemodule platform (RMP). There is no directcommunication between application mod-ules. All communication between themtakes place via protocols, where the proto-col carriers are implemented in the resourcemodule platform (Figure 1).

APZ

XSSAM

CONRM

COMRM

OMRM

APSI

NAPRM

RMP

OtherRMs

AM AM AM

CHARM

C7RM

Figure 1AXE 810 application software architecture.

The interface of the resource module plat-form with the application modules, calledthe application platform service interface(APSI), is a formalized client-server inter-face. Subsets of the APSI, or individual in-terfaces, are called service specifications(SES). Each service specification provides aservice for the application modules, some ofthe most important services being: connection servicefor setting up a phys-

ical connection; communication servicefor communi-

cation between application modules; object managerfor operation and main-

tenance; AM protocol carrier (APC); charging service; and services for signaling no. 7 (SS7), such as

the message transfer part service.The total concept, which consists of the

RMP, APSI, the AMs, and the inter-AMprotocols, is known as the application mod-ularity concept. It is largely thanks to thisconcept that the system lifetime of AXEconsistently exceeds expectationstheAXE system is 30 years old, yet we see noend to its potential evolution.

Resource modules

When the RMP was first introduced, it wassmall in comparison to the large volume ofexisting software, called existing source sys-tem (XSS). Following several RMP devel-opment projects, the migration from theearlier architecture to the AM architectureis now virtually complete. The resourcemodule platform, in turn, has become largeand complex, leading to a refinement of theAM concept and the division of the RMPinto several resource modules (RM). The in-terfaces provided by the resource modulesare formalized as resource-layer service spec-ifications (RLSES).

A priority in the development of RMs andAMs (known collectively as system mod-ules) is to specify the interfaces (SES andRLSES) early in the development process.When the interfaces are frozen, the sepa-rate system modules can be designed inde-pendently of one another, often at differentgeographical locations, as is common in theEricsson organization.

The second major advantage for develop-ment of applications is that application-platform hardware is now associated withspecific resource modules and controlled bythem. Application modules simply requestservices via the APSI. Hardware is ownedby the software in the respective resourcemodules.

HardwareThe hardware (Figure 2) revolves aroundthe GS890 group switch, which has 512 Kmultiple positions, each with a bit rate of64 kbit/s.

The biggest change in the hardware ar-chitecture is the addition of a new exchangeinterface (the DL34) to existing DL3 andDL2 interfaces. The DL34 interface sup-ports from 128 to 2,688 timeslots to eachdevice board, in steps of 128, via a backplaneinterface in the generic Ericsson magazinerunning at 222 Mbit/s. This interface madeit possible to improve the devices connect-ed to the group switch, further reducinginput power, cabling, the number of boardtypes, footprints, installation time, andother parameters. The high-speed DL34 in-


SCB-RP

ET155

ET155

XDB DLEB

DLHBTRA

ECP

CGB

CGB

SCB-RP

RPHS

CPHD

RPHS

APG

RP4

IRB

IRB

LRB

ETC5

ETC-T1

ETC-J32

RPP

RPG3

PDSPL2

J7X21-2

DL2IODL2 (cable)

VGA display

External alarm

Remote Ethernet (TCP/IP)

RPB-S

RPB-SEthernet

RPB-SEthernet (IPN)Ethernet (IPN)

Local Ethernet (TCP/IP)

Keyboard Alarm panel

ExternalsyncETCsync

SONET

EthernetRPB-S

GEM GDM

EMBUSRPB-S

DL3

4

DL3

4

DL3

DL2

SDH

SDH

PDH

PDH

PDH

ALI

M-AST

S7V35

S7DS0A

Figure 2Hardware architectural overview.

Connection RM (CONRM) Communication RM (COMRM) Operation and maintenance RM (OMRM) Network access products RM (NAPRM) Pooled devices RM (PDRM) Common channel signaling RM (C7RM) Charging support RM (CHARM)

BOX B, CURRENT RESOURCEMODULES


terface has also facilitated a more efficientversion of the group switch itself, also lo-cated in the GEM.

Another architectural change (Figure 2)is the interplatform network (IPN). In thefirst phase, the IPN will be used to speed upcommunication between the central proces-sor (CP) and adjunct processors (AP). Theinterface is based on fast Ethernet. In a sec-ond phase, it will be upgraded to Gigabitspeed.

All devices that support the DL34 inter-face can be mixed in the GEM more or lesswithout limit. The maximum capacity ofeach GEM is 16 K multiple-position time-slots (MUP), which corresponds to eightSTM-1 ET boards, for example. Physically,there is space for 22 device boards in oneGEM.

If a switch is needed that is larger than 16 K MUPs, or when the number of devicesexceeds 22, additional GEMs must beadded. These can be configured without in-terruption while the system is processingtraffic. An exchange based on the GEM islinearly expandable. The maximum switchsize is 512 K MUPs at normal rate (64kbit/s) or 128 K MUPs at a subordinate rate (n x 8 kbit/s).

By inserting a pair of digital link multi-plexer boards (DLEB) in the GEM, we canconvert the DL34 backplane interface intofour DL3 cable interfaces. This allows all oftodays devices in the generic device maga-zine (GDM), generic datacom device mag-azine (GDDM), and other formats to be usedwith the new switch core.

Software evolutionThe original AXE concept has enabled on-going evolution and modernization. Legacysoftware can be reused or modified, sub-systems and system modules can be mod-ernized or completely replaced, interfacescan be left unchanged or redesigned to meetnew needs, and new architectures can be in-troduced. Yet AXE remains fundamentallyAXE.

Several software development programsare under way in parallel. The APZ proces-sor is being adapted to serve as a multi-processor CP to multiply call-handling ca-pacity. Cost of ownership is being reduced.Ericsson is improving its internal manage-ment of software to reduce time-to-marketand time-to-customer. These activities arein addition to the four programs describedbelow.

Software migrationThe network architecture is changing. Inparticular, we see two new types of node: theserver (or media-gateway controller) and themedia gateway itself, resulting in what isknown as the server-gateway split. Inorder to meet the demands of the new ar-chitecture, the software in the applicationplatform is being migrated from a tradi-tional telecom environmentwhich isSTM-based, mainly for voice trafficto-

XSS

AM

RMP Server module

STM AAL2/AAL5

SS7SS7

GCP

GCPSS7 STM

1999 2001 2001/2002

STM AAL2/AAL5

Gateway module

Server module

Media gateway

AM AM AM AM AM

Figure 3Software evolution toward the server-gateway split.

ET

TRAor DTI

TRAor DTI

ALI

ATM/AAL2network

STM network

GSS

AXE media gateway

Switched 64 kbit/s connections for PCM coded voiceSwitched 64 kbit/s connections for HDLC framed voice or data packets

Figure 4System architecture, ATM interworking function.

ward a packet-based environment, initiallyfor ATM, and subsequently for IP.

In this new architecture, the server con-trols the call or session, whereas the mediagateway provides the bearer service, trans-mitting information over the connectivitylayer. An advantage of separating call con-trol from bearer control is that bearer tech-nology can be upgraded from STM via ATMto IP without affecting the call control.

The original AM architecture has proveditself to be future-proof and is well suited toaccommodate the server-gateway split. Thesplit has been implemented in system mod-ules (new and modified AMs and RMs)without any fundamental changes to thesoftware architecture. For example, newRMs for bearer-access service and for TDMaccess are currently being developed for theWCDMA/UMTS program.

Two separate migration tracks are currently being followed. For theWCDMA/UMTS program, AXE is the nat-ural choice of technology for the server,thanks to its high capacity and high level offunctionality. AXE is also available as a com-bined server/media-gateway node, where theserver part controls its own gateway part. Bycontrast, the Cello packet platform (CPP) isthe preferred choice of technology for ATM-based media gateways. In such a network,the server nodes communicate using bearer-independent call control (BICC). The servercontrols the gateway using the gateway con-trol protocol (GCP). ATM-based gatewayscommunicate using Q.AAL2.

The second migration track is for the next-generation switch (NGS) program for themultiservice network. The server node is ahybrid node made up of co-located AXE andAXD 301 systems. The gateway node con-sists of an ATM switch (also an AXD 301) for AAL1 connections, and iscontrolled by the server using the gatewaycontrol protocol.

In-service performanceRobustness, or in-service performance, haslong been a priority in AXE development,and considerable improvements have beenand are still being made. No software is100% reliable, but robust systems can beachieved by means of recovery mechanismsthat minimize the effects of software mal-functions or program exceptions.

One of the most powerful robustnessmechanisms is known as forlopp, which is arecovery mechanism at the transaction level.(Forlopp is an anglicization of the Swedish

word frlopp, roughly the sequence ofevents.) The forlopp mechanism is a low-level recovery mechanism by which only thetransaction affected by the software mal-function is released. This minimizes the dis-turbance to the overall system. The forloppmechanism, which has been refined in sev-eral stages and is now applied to all traffichandling in AXE and to many operation andmaintenance functions, has significantly re-duced the number of recoveries at the sys-tem level.

System restart is used as a recovery mech-anism for certain faults and for system up-grades. The restart mechanism itself hasbeen optimized in several ways, so that inthe few cases that still require restart, its du-ration is as short as possible.

Activities for restarting software in the re-gional processor (RP) have been especiallyimproved. For example, regional processorsare restarted through minimal restart by de-faultthat is, with the suppression of un-necessary actions. Complete restarts are per-formed on regional processors only whennecessary. Regional processors are restartedasynchronously, which means that no re-gional processor has to wait for any other re-gional processor to become ready for a restartphase. These improvements have signifi-cantly reduced the time consumed whenrestarting application hardware.

If necessary, AXE can, via the IPN, be re-loaded from the backup copy of the systemon the APG40 file system. This approach isten times faster than the design that appliedbefore the APG40 and IPN were intro-duced. Also, the time it takes to make abackup copy of the system on APG40 is one-fifth of what it was before.

The restart duration for a system that isupgrading to new software has also been re-duced. The most recent improvements in-volve RP software, which is now preloadedprior to an upgrade, instead of during theupgrade, thus saving restart time.

Major improvements have been made tothe function-change mechanism used forsystem upgrades. Inside the central proces-sor, the time needed for data transfer priorto a changeover to new software has been cutfrom hours to typically ten minutes. (Thedata transfer does not disturb traffic han-dling, but exchange data cannot be changedduring data transfer.)

Another improvement applies to theretrofit of new CPs that replace old ones. Thebandwidth of the connection between theprocessors has been expanded by using re-


1999

System downtime

Complete exchange failureSoftware upgradeRestart with reloadLarge restartSmall restart

2000 2001 2002

Figure 5Reduction in system downtime.


gional processor with industry-standardPCI interface (RPP) and Ethernet connec-tions, thereby reducing data-transfer timesten fold. As a consequence of these many im-provements, system downtime has been re-duced significantly in recent years, a trendthat is sure to continue (Figure 5).

Software licensingThe main force driving software licensing iscustomer preferences to pay for access to spe-cific functions, features, and capacity. Thus,software licensing is really the licensingof features and capacity. The second dri-ving force is Ericsson itself, because it is inthe companys interest to deliver standardnodes (those that contain standardized soft-ware and hardware).

With software licensing, Ericsson deliv-ers a standard node for a particular marketor market segment, such as an MSC for a cel-lular system using time-division multipleaccess (TDMA). This standard node consistsof a complete software configuration with astandard hardware configuration that is de-liberately over-dimensioned. Ericsson thenlimits the call capacity and functionality ofthe node by means of software keys. Whena customer requests more call capacity,Ericsson personnel execute a password-protected command to increase the capaci-ty in the node to the new level.

This method of increasing capacity, re-ferred to as traffic-based pricing or pay-as-you-grow, is much simpler than the tra-ditional method, in which Ericsson person-nel would deliver, install, and test new hard-ware on a live node. The commands can alsobe executed remotely from a maintenancecenter, making a visit to the site unneces-sary. This arrangement benefits customersand Ericsson, especially in the mobile mar-ket where growth can be rapid and unpre-dictable.

Similarly, in the future, software will con-tain functions and features for which cus-tomers can purchase licenses on an ad hocbasis. The software for these functions andfeatures will be unlocked using password-protected commands. This method of de-livering software is much simpler than thetraditional method, by which new softwareis delivered in the form of software upgrades.

The licensing of call capacity is alreadyavailable on AXE nodes, and the introduc-tion of a general mechanism to handle allsoftware licensing of functionality and ca-pacity is planned. Some of the characteris-tics of this general mechanism are as follows:

Software licensing will be common toAXE 810, TSP, CPP, and other Ericssonplatforms.

Ericsson will maintain a central licenseregister, from which new licenses can eas-ily be issued.

License keys will be distributed electron-ically, in encrypted form.

Improved handlingSeveral improvements to the operationalhandling of AXE have recently been made,such as parameter handling and hardwareinventory. Other handling improvementsare being planned, such as plug-and-playfunctionality. These improvements reduceoperational costs and also often have a pos-itive impact on the ISP, since they reducethe risk of human error.

One of the most important improve-ments to handling is called dynamic sizealteration (DSA). The size of many datarecords in AXE is traffic-related, since thenumber of individuals, or instances, in therecord varies from node to node, as well asover time, in accordance with the capaci-ty demands put on the node. Thanks todynamic size alteration, the number of in-dividuals in the record is increased auto-matically, up to a preset limit, withoutany intervention from the maintenancetechnician. However, as a warning, AXEissues an alarm when, say, 75% of the re-served data capacity has been used up, sothat the technician can raise the limit asnecessary.

Dynamic size alteration also simplifies thehandling of this kind of alteration, since thealarm (referred to above) specifies an actionlist of the data records whose size needs tobe altered (increased). The technician mere-ly enters a single command to increase thesizes of all these data records.

Hardware evolutionGeneric Ericsson magazineAnother far-reaching improvement is theGEM, a high-capacity, flexible, and scalablemagazine (subrack) that anticipates futuredevelopments (Figure 6). GEM-based nodeswill be smaller, dissipate less power, andhave greater maximum capacity. Their im-plementation will dramatically improve thecost of ownership and cut time-to-customerfor AXE.

In previous versions of AXE, each func-tion was located in a separate magazine, and

Figure 6The GEM, a high-capacity, flexible andscalable magazine (subrack).

the associated regional processors were onseparate boards in the magazine. The GEMand the boards developed for the GEM rep-resent fundamental change.

Several functions are mixed in a singlemagazine, and the RP function is integrat-ed on each board. This makes the most of

advances in technology to modify the archi-tecture rather than merely shrink the hard-ware. One advantage of this change is thatconsiderably fewer magazine and boardtypes are needed in each node. A second ad-vantage is that many internal AXE inter-faces have been moved to the GEM back-plane, reducing the need for cables. Asshown in Figure 7, many different functionscan be mixed in the GEM.

Boards developed for the GEM have highcapacity, and because more functions areconcentrated on each board, the total num-ber of boards can be reduced. Mixing thesehigh-capacity boards in a single magazineand removing many cables by moving theinterfaces to the GEM backplane producesa more compact system.

The GEM concept can be used to buildanything from an extremely small node toan extremely large one, using the sameboards and magazine (Figure 8). The small-est node will consist of one GEM, and thelargest, 32 GEMs. As capacity requirementsgrow, the node can be expanded smoothlyand without interruption of traffic byadding one or more GEMs with a suitablecombination of devices. A GEM can house two SCB-RPs, providing an Ethernet

switch, control function, maintenancesupport, and power distribution;

two switch boards (XDB), providing a 16 K plane duplicated group switch;

up to 22 device boards with 15 mm spac-ing, such as ET155, ECP, or TRA;

pairs of DLEBs, providing multiplexingfunctionality for DL3 cable interfaces,placed in device slots; and

CL890 clock modules placed in deviceslots.The GEM, which has been developed for

use in the BYB 501 equipment practice,provides several infrastructure functions viathe backplane for the boards housed in themagazine: duplicated power distribution; duplicated group switch connection; duplicated serial RP bus; maintenance bus for hardware inventory

and fault indication; duplicated Ethernet, 10 or 100 Mbit/s;

and extremely robust backplane connectors.The option to use Ethernet as the controlprotocol makes the GEM an open magazineprepared for future developments. Anotheradvantage of the GEM is that it is genericand can be used in non-AXE products fromEricsson.


Switch magazine GS multiplexormagazine Clock magazine

ET155 magazine Echo cancellermagazine Transcoder magazine

Figure 7Many different functions can be mixed in the GEM.

Ethernet to/from RPI

Ethernet to/from CP

RBS-S to/from RPI New group switch New devices RPI on each circuit board

RBS-S to/from CP

RPIRPI

GS890CL890

RPI

RPI

RPI

RPI

RPI

TRA

ET155-1

ECP5

DLEB

SCB-RP

Figure 8The GEM concept can be used to build anything from an extremely small node to anextremely large one, using the same boards and magazine.


The GS890 group switchThe GS890 fully supports the GEM conceptand its goal of dramatically improving costof ownership and time-to-customer. Thefigures in Table 1 show an extraordinary re-duction in power dissipation and number ofboards and cables. This was achievedthrough a combination of recent advances inASIC technology and high-speed interfaceswith architectural modifications.

The maximum size of the GS890 has beenincreased considerably, making it possibleto build larger nodes. For many networkconfigurations, it would be better to buildand maintain a small number of large nodes.The switch is also strictly non-blocking, somany limitations on the configuration of thenode and the surrounding network havebeen removed.

The subrate switch function in AXE,which enables switching at 8 kbit/s, has alsobeen modified. Primarily used in GSM net-works, this function makes efficient use ofpooled transcoders and transmission re-sources surrounding the base station con-troller (BSC). The subrate function was pre-viously implemented as a pooled function,but in GS890 (Box C) it has been integrat-ed into the switch core and increased to 128 K. The advantages are that much larg-er BSCs are feasible, less equipment is need-ed, and the traffic selection algorithms areeasier and faster because the pooled switchhas been removed.

GS890 group switch connection

All existing group switch interfaces are sup-ported, making it possible to connect allcurrent devices to the GS890. The DL2,which is a cable and backplane interface, car-ries 32 timeslots. Similarly, the DL3, whichis a cable interface used by the ATM inter-

working unit (IWU) and GDM, carries 512timeslots. The new DL34 backplane inter-face supports new high-capacity devices de-veloped for the GEM.

GS890 hardware

The main functionality of the GS890 is im-plemented in two new ASICs: one coreASIC, which handles the switching; and onemultiplexer ASIC, which concentrates traf-fic into the core. Both ASICs have been de-signed using 0.18 m CMOS technology.

The core ASIC contains 4 Mbit of RAM,runs at up to 333 MHz, has 500 K gate logicand 20 high-speed interfaces running at 666 Mbit/s, allowing the ASIC to handle upto 13 Gigabits of data every second. In a fullyexpanded switch matrix, 64 of these ASICswill be needed per plane. Standard four-paircable is used to interconnect the switch ma-trix elements located in different magazines.The cable can be up to 20 meters long andcarries 1.3 Gigabits of bidirectional data persecond, two 666 Mbit/s interfaces in each

Characteristics GS12 (Max 128 K) GS890 (Max 512 K) Reduction

Power dissipation in 1200 W 250 W 81%128 K MUP group switch core

Internal cables in a 1,024 (12-pair cables) 88 (4-pair cables) 92%128 K MUP group switch core, and 48 (4-pair cables)and including clockdistribution

Number of boards in a 320 16 95%128 K MUP group switch core,excluding clock boards

Equivalent figures 1,000 W, 448 (4-pair for a 512 K group switch cables) and 64 boards

TABLE 1, IMPROVEMENTS IN THE GROUP SWITCH

Maximum size 512 K MUP, equivalent to524,288 (64 kbit/s) ports

Strictly non-blocking architecture, regardlessof traffic type

Fully distributed group switch, with switchmatrix distributed among up to 32 GEMs

Integrated subrate switching capability up to128 K MUP, equivalent to 1,048,576 (8 kbit/s)ports

Improved maintenance support by logic BISTand RAM BIST in all ASICs, reducing mainte-

nance time and the time that the system runson one plane

Hardware support for fast dynamic fault iso-lation

New and flexible high-speed-device interface Device protection support with no wasted

capacity (used for ET155-1) New high-speed, internal group switch inter-

face Rear cabling removed

BOX C, MAIN CHARACTERISTICS OF THE GS890

direction. The new hardware requires twonew board types: the switch board; and the multiplexing board, which makes it

possible to connect existing devices. In addition, two more boards have been de-veloped to support a smooth migration fromthe GS12 to the GS890. The migration con-sists of replacing one board in each GS mag-azine, thereby converting it into a multiplexing magazine. The concentratedtraffic is then connected to the new GS890core. These boards help safeguard invest-ments already made in AXE. The GS890(Figure 9) supports the following types oftraffic: normal rate, 64 kbit/s; non-contiguous wideband, n x 64 kbit/s

up to 2 Mbit/s;

subrate, 8 kbit/s or 16 kbit/s; wideband on subrate, n x 8 kbit/s up to

256 kbit/s; and broadband connections up to 8 Mbit/s.The CL890 synchronization equipment,also developed to support the GEM concept,consists of duplicated clock modules; up to two highly accurate reference clock

modules; and up to two incoming clock reference boards

for connecting additional clock refer-ences.

Devices

RPG3

A new version of the RPG2 with slightlyimproved CPU performance has been de-signed that only occupies one slot instead oftwo. The RPG3, which is still located in theGDM, is mainly used for different kinds ofsignaling application.

ET155-1

The ET155-1 is an SDH exchange terminalthat will occupy one slot in the GEM andwill be available for ETSI andANSI/SONET standards. The ETSI variantwill carry up to 63 TU12s and support bothoptical and electrical line interfaces. InANSI/SONET mode, the board will be ableto handle up to 84 TU11s.

The board fully supports channel-associated signaling (CAS) in both standardsby extracting the signaling informationfrom the incoming PDH frames and send-ing it through the switch to pooled signal-ing terminals. The ET155-1 can work ei-ther as a non-redundant single-board ex-change terminal or in tandem, supportingMSP 1+1 and equipment protection.

One motherboard, which contains themain functionality, and two generic line in-terface modules, containing the optical re-spectively electrical line interface, have beendeveloped. An ET155-1 board is producedby mounting the required module on themotherboard. The line interface modules aregeneric and will also be used in non-AXEproducts from Ericsson.

Transcoders and echo cancellers

A big improvement has been made in thetranscoder area. A new board designed for theGEM supports 192 channels for all the codecsused in GSM and TDMA systems. The sameboard is also used for echo cancellation, inwhich case it supports 128 channels.


SDH ATMSTM-1

DL3 to group switch

RPB-S to CP

AAL2 GSinterface

AAL5 RP

Figure 10Block diagram of the ATM link interface(ALI).

Figure 9Photograph of the switch distributedboard (XDB).


A fully equipped GEM with transcoderssupports 4,224 transcoder channels; whenfully equipped with echo cancellers, theGEM handles 2,816 channels.

Future GEM devices

The devices described here were designedfor the first release of the GEM. Future re-leases will include more devices ported fromtodays GDM and GDDM. Technology al-lows more traffic to be handled by each boardin AXE, so the boards can efficiently takeadvantage of the configurable bandwidth ofthe DL34.

ATM interworking functionWith the introduction of WCDMA/UMTS,AXE will be used as a combined server andmedia gateway and purely as a server. In thedownlink direction (to the radio access net-work), AXE will be attached to the Iu in-terface, a new interface based on ATM. ATMwill also be used in the core networks.

An ATM interworking unit has been im-plemented to add an ATM interface to newand legacy AXE nodes. The interworkingunit enables existing AXE-based MSCs tobe upgraded easily to handle signaling andthe user plane in ATM networks. The inter-working unit is connected directly to thegroup switch interface for the user plane,and the signaling information is transferredto the central processor via the RP bus (Fig-ure 10).

The interworking unit provides a 155 Mbit/s ATM/SDH interface to AXE.New hardware and software for AXE enableit to perform as a gateway to an ATM net-

work capable of handling voice, data, andsignaling. The first release of the ATM in-terworking unit supports three types of pay-load in the ATM network: coded voice on AAL2; circuit-switched data on AAL2; and signaling on AAL5.The device that implements the interworkingunit is called the ATM link interface (ALI).This hardware unit is divided into threeboards that are mounted together on a plug-in unit which can be located in a full-heightGDM that provides RP bus, 48 V and main-tenance bus via the backplane. The unit is thenconnected via DL3 cables to the group switch.The fiber for the STM-1 interface is connect-ed to the external equipment.

Adjunct processorplatformStatistics from the telephony or radio parts ofa network must be collected to optimize net-work efficiency. At the same time, virtual op-erators are willing to pay for the right to usetelecom equipment owned by other opera-tors, and network operators are demandingfast, almost instant billing (such as with pre-paid cards). These requirements demandprocesses with close to real-time execution.

In terms of operation and maintenance, itis better to have one big node in a networkinstead of several small ones, although thisdemands greater switching capacity. To re-duce operators cost per subscriber, bigswitches are being planned. This has spurredthe development of new high-end AXEswitches. There are numerous methods of

To transmit coded voice over ATM, a perma-nent virtual circuit (PVC) is set up between theALI and the remote end. The PVC is specifiedby its VPI/VCI address. The first ALI releaseallows a total of eight PVCs to be set up in theuser plane.

Within each PVC, AAL2 connections (basedon the I.363.2 standard) can be set up andreleased using Q.2630.1 (Q.AAL2) signaling. Upto 248 AAL2 connections can be set up in eachPVC.

The C7 signaling links for supporting Q.AAL2and other user parts are configured as anoth-er type of PVC. For each ALI, up to 64 signal-ing PVCs can be configured to carry adapta-tion layer AAL5, based on the I.363.5 standard.Each of these AAL5 channels carries a C7 sig-

naling link using SSCOP (Q.2110) as the linkprotocol.

Adaptation layers for both ETSI (SSCF-NNI,Q.2140) and Japanese C7 standards (JT-Q.2140) are provided.

The architecture for supporting circuit-switched data is similar to that for coded voiceexcept that the payload packets are routed toa data transmission interworking (DTI) unitinstead of a transcoder. The first release of ALIsupports a total of 32 data channels.

In the future, the ATM interworking conceptmight be used to implement new functions suchas 64 kbit/s voice on AAL1 or on AAL2 basedon I.366.2. Another function in the works is softPVCs, which can be operated using B-ISUPsignaling.

BOX D, ATM VIA AXE

enhancing performance, since AXE pro-cessing power is distributed among severalprocessors (CP, RP, and AP), and researchand development can be pursued along dif-ferent lines in parallel.

APG40 platform for near real-timeapplicationsThe most widely deployed AXE IO system,IOG20, was introduced in 1997. TheIOG20 is a proprietary product for the CPfile system, man-machine communication,boot server, and statistics functionality. In

1998, the first adjunct processortheAPG30was released, primarily as abilling system. The APG30 is widely de-ployed in TDMA systems as a platform forbilling but also for operation and mainte-nance of AXE as an IOG replacement. TheAPG30, which is based on a commercialcomputer from an external vendor, usesMIPS RISC microprocessors and NonStop-UX O/S.

The APG40, successor to both the IOG20 and APG30, is a platform for nearreal-time applications. Three needs havedriven the development of the APG40 (Fig-ure 11). One need is to offer a platform fornew data-handling tasks and applicationsthat are best performed at the network ele-ment level by a standard computer system.Another need is for improved performance,as a result of increased traffic per switch andnew types of data extracted from eachswitch. The third need is to reduce time-to-market and time-to-customer.

The APG40 is a platform for the AXEcentral processor IO functions (APIO) thatwere inherited from the IOGs. It is also aplatform for billing (for example, FOS) andstatistic data collecting (for instance, STS),storage, processing and output from theAXE switch. For example, the APG40 canbe a platform for collecting data related toin-service performance from the centralprocessor. It can also format that data fordistribution to an operations management


Figure 11Photograph of a large APG40 cluster.

Proprietary middleware

FOS TMFSTSAPIO

Development environment

MSCS

CP-com

3pMW

Windows NT/2000

Hardware BYB 501, cPCI, Intel, RAID

TCP/IP

Figure 12APG40 building blocks and developmentenvironment.


center, such as the operations support sys-tem (OSS), for preventive measures or forsending an alarm when a critical level isreached.

The applications that run on the APG40do not communicate with one another andare not interdependent. The platform pro-vides basic functionality, such as highlyavailable processing capacity, safe storage,and data output mechanisms.

The adjunct computer subsystem (ACS)(proprietary middleware in Figure 12) pro-vides the software platform for applicationsin the APG40. The ACS offers WindowsNT, Microsoft cluster server (MSCS), andcommunication protocols, such as FTP,RPC, TCP/IP, and Telnet.

In-service performanceOperators require reliable input-output sys-tems that store AXE system backup copysafely and perform reloads fast. A switchmust always be available for operation andmanagement from an operations supportsystem. To ensure the in-service perfor-mance of the APG40 itself, and AXE as awhole, concepts introduced in the IOGs andAPG30 have been brought to this new plat-form. In the APG40, mainstream technolo-gies, such as MSCS and RAID hardware,form the basis for telecommunications-grade in-service performance.

The APG40 is a cluster server (Figure 13).In its first release, a cluster will consist of

two nodes. Each node has its own memoryand runs the operating system. All applica-tions running on an APG40 cluster are ac-tive on one node onlythe executive node;the other node is a standby node. Thus, if amajor fault occurs, the standby node startsand takes over the application. A watchdogin the MSCS ensures that the adjunct proces-sor software does not simply stop working.When an error is detected, either in hard-ware or software, the cluster server canrestart the failed resources in the other node.This process is called failover.

The physical size of the APG40 is a halfmagazine for a small cluster and one fullmagazine for a large cluster. The small APG40 has one 18 GB disk per node, whilethe large APG40 can house up to three 18 GB disks per node. Later on, it will bepossible to introduce disks with a capacityof 36, 72, or 144 GB.

Thanks to the RAID hardware system(Figure 13), the APG40 has a higher disk-integration speed than the IOG20. TheRAID hardware system is used for safe stor-age. Only the executive node can write datato the data disk. To avoid loss of data due toa disk error, the executive node writes datato a Windows NT file system, and the RAIDsystem writes to the data disk in both theexecutive node and the standby node. Theexecutive node thus owns all data disks inboth nodes. In the IOG20, on the otherhand, each node owns its own set of data

AXE internal Ethernet

External Ethernet

AXE internal Ethernet

APG40 cluster

Heartbeat

Node A

RAID

DataSystem System

CPU

Data

RAID

Node B

CPU

Failover

Figure 13The RAID hardware system.

disks, which must be re-integrated eachtime a failover occurs. Re-integration in theAPG40 occurs when a node is replaced orthe RAID system fails.

The AXE IO platforms listed in Table 2are based on three different high-availability concepts. In the IOG20, the pas-sive node is hot-standby, whereas in the APG40, the passive node is warm-standby.In this context, the main difference betweenhot- and warm-standby is that a larger partof the software is up and running in hotstandby. In the APG30with its fault-tolerant conceptidentical processes areexecuted in parallel on separate processorunits. If a hardware failure occurs, an im-mediate failover is initiated from the failingprocessor unit.

Behind the different availability conceptsin Table 2 lie different assumptions aboutthe most probable source of failure. If a hard-ware failure happens more often than a soft-ware failure, the fault-tolerant technologygives better in-service performance. Afailover in the APG30 executes in an instant,because the same processes, including theapplications, are running in parallel on bothsides.

Nevertheless, because the risk of hardwarefailure is considered extremely low, the casefor dimensioning has been based on softwarefailure, and the high-availability conceptused in the APG40 has been introduced asthe preferred solution. Given the risk ofhardware faults, application software faultsand system software faults, the failover peri-od has been held to a mininum. On theAPG40, these three kinds of fault are wellbelow the minimum CP buffer capacity oftwo minutes, which means that no data islost.

Performance and functionalityenhancementsThe demands put on the adjunct processorsare not the same as those put on the centralprocessor. The central processor providesnon-stop computing, so the adjunct proces-sor concept is based on and provides high-availability computing. Consequently, newprocessor systems can be obtained frommainstream data communications vendors.The first generation APG40 is based on IntelPentium III 500 MHz processors and Mi-crosoft NT4 clustering technology. Byadapting these mainstream products to spe-

Ericsson Review No. 1, 2001

APZ 212 25

APG40

APZ212 30

GS890TRA

ET155

GS890TRA

ET155

GS890TRA

ET155RPP

RPP

RPP

GS890TRA

ET155

RPPRPG3/RPP

RPG3

GS890ET155-1

APZ212 30 APG40

RPG3RPG3RPG3

PowerEDU

1.8

m

0.6m

1.8m

0.4m

0.8m

BSC500TRX

BSC/TRC1020TRX

Figure 14Node layout for BSC 500 TRX andBSC/TRC 1020 TRX.

MIPS is a registered trademark of MIPS Tech-nologies, Incorporated.NonStop-UX is a trademark of Compaq Infor-mation Technologies Group, L.P. in the Unit-ed States and/or other countries.Windows and Windows NT are registeredtrademarks of Microsoft Corporation.Intel and Pentium are trademarks or registeredtrademarks of Intel Corporation.Rational and ClearCase are trademarks or reg-istered trademarks of Rational Software Cor-poration in the US and in other countries.

TRADEMARKS


cific needs, Ericsson benefits from ongoingresearch and development conducted by oth-ers. This approach also simplifies the rapidintroduction of new features and applica-tions using Microsoft and Rational develop-ment tools integrated in an adjunct proces-sor design environment. Examples of this aredebugger and Rational test tools integratedin Microsoft Visual C++, RationalClearCase, and HTML online help.

Other examples of mainstream technolo-gy being introduced full-scale in the AXEby the APG40 are the TCP/IP, FTP, andEthernet (for external communication andinternal AXE communication), and thecPCI building practice. By introducingTCP/IP and Ethernet in the IPN for CP-APcommunication, the central processor reloadspeed is 10 times that obtained when usingthe APG40 with signaling terminal opencommunication (STOC).

The introduction of TCP/IP as the exter-nal interface to the OSS is a major step to-ward replacing the propritary interfacesbased on MTP and X.25. The external Eth-ernet board will enable a TCP/IP connec-tion of up to 100 Mbit/s.

Telnet has been introduced for handlingcommands. Similarly, FTP and RPC areused for handling files. Support for a com-ponent object model-based (COM) OSSmight be introduced in the future.

For local element management of theAXE 810, WinFIOL and Tools 6.0 (andlater versions) have been adapted to the newAPG40.

ResultsBest-in-class hardware embodies compactlayouts and easy extensions, exemplified inFigure 14 by one BSC and one combinedBSC/TRC for the GSM system.

The BSC, which has a capacity of 500transceivers, occupies only one cabinet. Atthe top, the CP (APZ 212 25) and APG40 share one shelf. Below this are threemagazines with a mix of RPPs and RPG3s.The C7, transceiver handler and PCU ap-plications run on this hardware. At the bot-tom of the cabinet, one GEM contains ET155, synchronization, DL3 interfaceboards (DLEB) and the group switch.

The combined BSC/TRC has a capacity of1,020 transceivers and approximately 6,000Erlang. The APZ 212 30 is needed for thislarger node, occupying two cabinets. An-other cabinet houses four GEMs, each of which contains a mix of transcoders,

ET155, synchronization, DL3 interfaceboards and the group switch. The next cabinet contains one APG40, the RPG3s(for transceiver handling) and C7. Thepower equipment is located at the bottom.PCUs can be included in the last cabinet ifthe BSC/TRC is intended to handle GPRStraffic.

ConclusionThe AXE architecture allows us to continueto improve the different parts of the systemas required by network evolution. This en-sures that operators always have a networkwith the best possible cost-of-ownership. Alimited number of high-capacity nodes witha limited footprint keep site costs down.State-of-the-art hardware with reduced ca-bling also ensures quick, high-quality in-stallation procedures. Operation and mainte-nance is kept economical through low powerconsumption, an interface consistent with theOSS, and a high-capacity, user-friendly extension-module platform. The software architecture allows migration toward next-generation networks, whether fixed orwireless.

1 Ekeroth, L. and Hedstrm, P-M.: GPRS sup-port nodes. Ericsson Review Vol. 77(2000):3, pp. 156-169.

2 Hgg, U. and Lundqvist, T.: AXE hardwareevolution. Ericsson Review, Vol. 74 (1997):2, pp. 52-61.

3 Hallenstl, M., Thune, U. and ster, G.ENGINE server network. Ericsson ReviewVol. 77 (2000):3, pp. 126-135.

4 Andersson, J-O. and Linder, P. ENGINEAccess RampThe next-generation accessarchitecture. Ericsson Review Vol. 77(2000):3, pp. 146-147.

REFERENCES

HSB FT HA(IOG20) (APG3x) (APG4x)

Hardware fault Software failover Hardware failover Software failover90 sec. < 1 sec. < 60 sec.

Application soft- Software failover Process restart Process restartware fault 90 sec. 22 sec. < 1 sec.

System software Software failover Reboot Software failoverfault (major) 90 sec. 80 sec. < 60 sec.

TABLE 2, THREE AXE IO SYSTEMS

ericsson

Documents

generation networks

generation of networks

data networks

fixed networks

general networks

existing networks

fortraditional networks

axe exchange