A Network Administrator's View of Multiservice Networks
This chapter covers the following topics:
- The Origins of Multiservice ATM
- Next-Generation Multiservice Networks
- Multiprotocol Label Switching Networks
- Cisco Next-Generation Multiservice Routers
- Multiservice Core and Edge Switching
Multiservice networks provide more than one distinct communications service type over the same physical infrastructure. Multiservice implies not only the existence of multiple traffic types within the network, but also the ability of a single network to support all of these applications without compromising quality of service (QoS) for any of them.
You find multiservice networks primarily in the domain of established service providers that are in the long-term business of providing wireline or wireless communication-networking solutions year after year. Characteristically, multiservice networks have a large local or long-distance voice constituency and are traditionally Asynchronous Transfer Mode (ATM) Layer 2-switched in the core with overlays of Layer 2 data and video solutions, such as circuit emulation, Frame Relay, Ethernet, Virtual Private Network (VPN), and other billed services. The initial definition for multiservice networks was a converged ATM and Frame Relay network supporting data in addition to circuit-based voice communications. Recently, next-generation multiservice networks have emerged, adding Ethernet, Layer 3 Internet Protocol (IP), VPNs, and Multiprotocol Label Switching (MPLS) services to the mix. IP and, perhaps more specifically, IP/MPLS core networks are taking center stage as multiservice networks are converging on Layer 2, Layer 3, and higher-layer services.
Many provider networks were built piecemeal—a voice network here, a Frame Relay network there, and an ATM network everywhere as a next-generation voice transporter and converged platform for multiple services. The demand explosion of Internet access in the 1990s sent many providers and operators scrambling to overlay IP capabilities, often creating another distinct infrastructure to operate and manage. Neither approach used the current investment to its best advantage.
This type of response to customer requirements perpetuates purpose-built networks. Purpose-built networks are not solely a negative venture. These networks do serve their purpose; however, their architectures often overserve their intended market, lack sufficient modularity and extensibility, and, thus, become too costly to operate in parallel over the long term. Multiple parallel networks can spawn duplicate and triplicate resources to provision, manage, and maintain. Examples are resource expansion through additional parts sparing, inimitable provisioning and management interfaces, and bandages to the billing systems. Often a new network infrastructure produces an entirely new division of the company, replicating several operational and business functions in its wake.
The new era of networking is based on increasing opportunity through service pull, rather than through a particular technology push requiring its own purpose-built network infrastructure. Positioning networks to support the service pull of IP while operationally converging multiple streams of voice, video, and IP-integrated data is the new direction of multiservice network architecture. In the face of competitive pressures and service substitution, not only are next-generation multiservice networks a fresh direction, they are an imperative passage through which to optimize investment and expense.
In this chapter, you learn why the industry initially converged around ATM; about next-generation multiservice network architectures that include Cisco multiservice ATM platforms, IP/MPLS routing and switching platforms, and multiservice provisioning platforms; and about multiservice applications that converge data, voice, and video.
The Origins of Multiservice ATM
In the early 1980s, the International Telecommunication Union Telecommunication Standardization sector (ITU-T) and other standards organizations, such as the ATM Forum, established a series of recommendations for the networking techniques required to implement an intelligent fiber-based network to solve public switched telephone network (PSTN) limitations of interoperability and internetwork timing and carry new services such as digital voice and data. The network was termed the Broadband Integrated Services Digital Network (B-ISDN). Several underlying standards were developed to meet the specifications of B-ISDN, including synchronous optical network (SONET) and Synchronous Digital Hierarchy (SDH) as the data transmission and multiplexing standards and ATM as the switching standard. By the mid-1990s, the specifications for the ATM standard were available for manufacturers.
Providers began to build out ATM core networks on which to migrate the PSTN and other private voice networks. Partly justified by this consolidation of the voice infrastructure, the ATM core was positioned as a meeting point and backbone carrier for the voice network products and the Frame Relay data networks. ATM networks were also seen as enablers of the growing demand for multimedia services. Designed from the ground up to provide multiple classes of service, ATM was purpose-built for simultaneous transport of circuit voice, circuit-based video, and synchronous data.
ATM was not initially designed for IP transport but rather was designed as a multipurpose, multiservice, QoS-aware communications platform. It was primarily intended for converging large voice networks, H.320 video networks, and large quantities of leased-line, synchronous, data-based services. ATM theory was heralded as the ultimate answer to potentially millions of PC-to-PC, personal videoconferencing opportunities. It was anticipated that its fixed, cell-based structure would be easily adaptable to any type of data service, and, indeed, adaptation layers were designed into ATM for transport of IP and for LAN emulation.
In essence, ATM was part of a new PSTN, a new centrally intelligent, deterministic pyramid of power that was expected to ride the multimedia craze to mass acceptance. As such, many service providers who needed a core upgrade during the 1990s chose ATM as a convergence platform and launch pad for future services.
ATM is a system built on intelligence in switches and networks. In contrast, IP-based products are built on intelligence in the core and intelligence distributed to the edges of networks, primarily in customer edge computers that summon or send data at their master's will. In fact, it is the bursty, variable, free-roaming data characteristics of IP that effectively cripple the efficiency of ATM for IP data transport.
Running IP packets through the ATM Adaptation Layers (AALs) creates a hefty overhead referred to as the ATM cell tax. For example, an IP packet of approximately 250 bytes will need to be chopped and diced into several 48-byte payloads (5-byte ATM header per cell for 53 total bytes), and the last cell will need to be padded to fill out the full data payload, the padding becoming extra overhead. A 250-byte IP packet using an AAL5 Subnetwork Access Protocol (SNAP) header, trailer, and padding swells to 288 bytes with a resulting cost of about 15.2 percent overhead per packet. The shorter the length of an IP packet, the larger the percentage of overhead. TCP/IP is all over the map—size wise—with many data packets, especially acknowledgements, shorter than 100 bytes. Using ATM Inverse Multiplexing over ATM to bond T1 circuits for a larger bandwidth pool in ATM networks imposes significant, additional overhead. Adding it all up, the total fixed and variable cell tax overhead can be decimating to linkage of IP traffic.
Back in the late 1990s when IP networks were coming on very strong, ATM products for enterprises cost about twice as much as Ethernet-based products, cost twice as much to maintain, and were intensive to configure and operate due to the ATM addressing structure and virtual circuit mesh dependencies. ATM was just too expensive to purchase and maintain (more tax) to extend to the desktop, where it could converge voice, video, and data.
ATM initially entered the WAN picture as the potential winner for multiple services of data, video, and voice. As with any new technology, the industry pundits overhyped the technology as the answer to every networking challenge within the provider, enterprise, and consumer markets. As IP networks continued to grow, and voice and video solutions were adapted to use IP over Fast and Gigabit Ethernet optical fiber spans, the relevance of ATM as a universal convergence technology waned.
Due to ATM's complexity of provisioning, its high cost of interfaces, and its inherent overhead, ATM gravitated to the niche bearers of complex skill sets, such as in service provider core networks, in large enterprise multiservice cores, and as occasional backbone infrastructure in LAN switching networks. ATM has also been a well-established core technology for traditional tandem voice operators and as backhaul for wireless network carriers. Much like ISDN before it, the technology push of ATM found a few vertical markets but only along paths of least resistance.
From a global network perspective, the ascendancy of IP traffic has served ATM notice. According to IDC, worldwide sales of ATM switches were down 21 percent in 2002, another 12 percent in 2003, and nearly 6 percent through 2004. Further, IDC forecasts the ATM switch market to decline at roughly 8 percent per year during the 2006 to 2009 timeframe.1
With the Digital Subscriber Line (DSL) deployments by the Incumbent Local Exchange Carriers (ILECs), ATM networks moved into the service provider edge, extending usefulness as broadband aggregation for the consumer markets. DSL has been an important anchor for ATM justification, bridging consumer computing to the Internet, but even there, DSL technology is signaling a shift to Ethernet and IP. The DSL Forum has presented one architecture that would aggregate DSL traffic at the IP layer using IP precedence for QoS rather than at the ATM layer. In Asia, many DSL providers already use Ethernet and IP as the aggregation layer for DSL networks, benefiting from the lower cost per bit for regional aggregation and transport.
Soon, ATM switching will likely be pushed out of the core of provider networks by MPLS networks that are better adapted to serve as scalable IP communications platforms. In fact, many providers have already converged their Frame Relay and ATM networks onto an MPLS core to reduce operational expenditures and strategically position capital expenditures for higher margin, IP-based services. ATM will settle in as a niche, edge service and eventually move into legacy support.
However, for providers that still have justifiable ATM requirements, there remains hope by applying next-generation multiservice architecture to ATM networks, which you learn about in the next section. Because providers cannot recklessly abandon their multiyear technology investments and installed customer service base, gradual migration to next-generation multiservice solutions is a key requirement. However, the bandwidth and services explosion within the metropolitan area, from 64 Kbps voice traffic to 10 Gigabit Ethernet traffic, is accelerating the service provider response to meet and collect on the opportunity.
Figure 3-1 shows a representative timeline of multiservice metropolitan bandwidth requirements. Through the 1980s and into the 1990s, bandwidth growth was relatively linear, because 64 Kbps circuits (digital signal zero or DS0) and DS1s (1.5 Mbps) and DS3s (45 Mbps) were able to address customer growth with Frame Relay and ATM services. The Internet and distributed computing rush of the late 1990s fueled customer requirements for Gigabit Ethernet services, accelerating into requirements for multigigabit services, higher-level SONET/SDH services, and storage services moving forward. The bandwidth growth opportunity of the last ten years is most evident in the metropolitan areas where multiservice networks are used.
Figure 3-1 Primary Metropolitan Traffic Timeline