Next gen ATM designs on existing platforms offer alternative to Ethernet MPLS

Indeed, because of the stringent requirements of low latency and jitter for carrier-class packetized voice, there is a renaissance in ATM technology and networks. While eventually, Multi-Protocol Label Switching (MPLS) may be the cure for IP's connectionless, best-effort limitations, the standardization process for MPLS is slow as the standards development organizations (SDOs) attempt to re-solve many of the issues that the ATM SDOs solved years ago.

It will also be some years before MPLS networks are widespread. Meanwhile, ATM standards and products are mature and ATM networks are ubiquitous. This view has been ratified by many key exciting next generation architectures propounded by groups such as the 3GPP, MSF, DSL Forum, ATM Forum, ETSI, and others.

Three next generation ATM architectures have emerged which show great promise and which can be mapped onto the current generation of commercial off-the-shelf platforms: UTMS 3 / 4, loop emulation using Services Using AAL2, and Next Generation Network Access (NGNA).

The specifications for the third generation cellular telephone network known as the Universal Mobil Telephone Service (UMTS) is largely a product of the Third Generation Partnership Project (3GPP). Understanding the QoS limitations of connectionless IP networks, the 3GPP specified ATM as the bearer plane for UMTS Releases 3 and 4. In particular, 3GPP specified the use of ATM AAL2 for the transport of the voice bearer streams.

In the UMTS model, the cellular base transceiver stations, known as NodeBs interface to the Radio Network Controller (RNC) via ATM AAL2. In turn, the RNC interfaces to other RNCs and to the circuit-switched core network through Media Gateways via ATM AAL2. And, for data services, the RNC interfaces to the packet core network via a Serving Gateway Services Node (SGSN) using ATM AAL5. In all cases, the subscriber's voice stream is carried by ATM AAL2.

Basically, AAL2 is an ATM Adaptation Layer specifically tailored for the transmission of compressed voice streams while still maintaining strict bounds on latency and jitter. It achieves this by packing multiple voice samples in a single ATM cell payload. By doing so, a consistent inter-sample gap can be guaranteed and thus jitter is minimized. However, this guarantee is only valid if the transport of the ATM cells is also done in strict accordance with ATM's service requirements.

Loop Emulation Services (LES) Using AAL2 is an evolving specification by the Voice and Multimedia Over ATM (VMOA) Working Group of the ATM Forum. The goal of this specification is to define the use of ATM AAL2 for the efficient and transparent transport of voice, modem, and fax streams to customer premise equipment across bandwidth-constrained links such as DSL.

Also known as Voice over DSL (VoDSL), this allows multiple standard analog phone lines to be emulated by ATM service across a single DSL link. Thus providers can offer small businesses multiple lines of voice service (up to 16) across a single pair of copper.

The LES specification defines a pair of Interworking Functions. The Customer Premise Interworking Function (CP-IWF) interfaces directly to both the customer's legacy telephone systems (PBX) and to a DSL line. The Central Office Interworking Function (CO-IWF) provides the interworking between an aggregation of the above subscriber streams (usually via a DSL Access Multiplexor or DSLAM) and the Central Office switch.

Third generation cellular telephone networks are implementing the Universal Mobile Telecommunications System (UTMS) model which uses ATM for voice delivery. The UMTS model uses NodeB interfaces to radio network controllers (RNC) via ATM AAL2. The RNC then interfaces to other RNSes and to the circuit-switched network through media gateways. Source: Motorola

Because both of these IWFs have to handle a variety of complex tasks such as voice activity detection, silence insertion, modem/fax tone detection and a variety of voice compression algorithms, the LES specification draws heavily on the ITU-T's Service Specific Convergence Sub layer for AAL2 specified in I.366.2.

Known as the AAL2 Trunking Specification, I.366.2 describes the procedures necessary to detect when one or both subscribers are silent and how to compress out the silence intervals to minimize bandwidth utilization. Because modem and fax streams don't survive the high compression algorithms that are meant specifically for human speech, I.366.2 also describes how the two IWFs can simultaneously shift to lower compression algorithms more friendly to those types of media.

Next Generation Network Access (NGNA) is another evolving specification coming out of a partnership between the VMOA Working Group of the ATM Forum and the European Telecommunications Standards Institute (ETSI) TIPHON Working Group. While the ETSI TIPHON (Telephony IP Harmonization Over Networks) Working Group is mostly concerned with Voice over IP and sees ATM as an access network into a TIPHON network, both bodies work closely to prevent NGN "islands" where subscribers to one form of service wouldn't be able to call subscribers in another form of NGN service.

Evolving networks
The goal of the ATM Forum NGNA specification is to enable the creation of a broadband access network that can deliver a wide range of voice, video, and conferencing services. The initial focus of this specification is on ATM networks but will evolve to other appropriate broadband networks in the future. The NGNA specification currently defines three network functions or elements: the Subscriber Media Gateway (SMG), the Transit Media Gateway (TMG), and the Media Gateway Controller (MGC).

The SMG essentially creates the bearer plane portion of a NG Central Office Switch. It provides the interface to the subscribers via standard analog phone lines, ISDN or the LES architecture.

As such, it is responsible for providing standard telephony services such as Call Waiting, Call Forwarding, Caller-ID, etc. and mapping those services to Voice over ATM (VoATM). The TMG provides the interworking necessary to access the classical long distance phone network or another Next Generation Network. The call control and bearer control necessary to connect an SMG with a TMG is provided by the Media Gateway Controller using H.248/Megaco signaling.

These three ATM applications are perhaps the hottest areas of development within the ATM aware SDOs. All of them leverage off of ATM AAL2's ability to support transmission of compressed speech and all of these architectures impose stringent QoS requirements on their underlying platforms. Implementing any one of the above network elements using current standards-based commercial off-the-shelf platforms poses a real challenge.

Support for embedded bearer-plane transport is currently limited in Compact PCI platforms to Ethernet ala the PCI Industrial Computer Manufacturers Group (PICMG) 2.16.

To build any of the network elements described using Ethernet as the backplane transport requires that the native ATM AAL2 streams be converted to UDP/IP streams upon entering the shelf. Then the voice streams would have to be converted back to ATM AAL2 upon leaving the shelf.

However, interworking ATM to UDP/IP/Ethernet just to support traffic distribution to the various payload cards within the shelf would destroy the QoS guarantees that mandated the use of ATM in the first place. This undesirable conversion step also places an otherwise unnecessary processing burden on many of the payload cards since each would have to be running an IP protocol stack.

In addition, the encapsulation of a 48-byte ATM cell payload in UDP/IP/Ethernet entails a significant loss of backplane bandwidth efficiency. The nine-byte ATM/AAL2 cell header would be traded in for a minimum 54-byte UDP/IP/Ethernet header on a one for one basis. This is a very poor use of backplane bandwidth when you consider that a typical compressed voice sample is just 20 bytes

Fortunately, there is a solution in the offing. The PICMG 2.20 proposed standard creates a fully meshed high-speed serial interconnect infrastructure which can be combined with a network processor card to implement a native ATM backplane infrastructure in a CPCI platform.

With such a backplane, ATM streams entering the shelf are transported as is to the various payload slots. After processing, these streams can exit the shelf still in their native ATM format. Looking forward, the PICMG3.x or Advanced TCA will support a similar backplane interconnect. With a backplane infrastructure that can support native ATM services, implementation of any of these next generation network elements becomes a practical and exciting possibility.