Industry pundits share a similar vision of future telecoms networks :integrated voice, video and data on a single data-centric network as just so much trafic.All broadly agree that circuit-switched voice networks will be gradually superseded by cell and packet switches and routers.
This vision only gets hazy when trying to pinpoint exactly when this transformation will occur - or even to say for certain that the multibillion dollar legacy in circuit-switched infrastructure will ever be completely replaced. Equally uncertain is the underlying question of whether ATM will continue as the dominant transport in core networks, or whether it will gradually be superseded by IP running directly over Sonet/SDH.
Telecom market research consultancy, Ryan Hankin, Kent Inc (RHK), of South San Francisco, says the migration from circuit-switching began in the US 'in earnest' this quarter, driven by growing Internet traffic and increasing bandwidth at the access network.
Sprint Communications is at the forefront of this change, having already started replacing its Class 4 circuit switches with ATM switches. Meanwhile, developments in voice-over-IP technologies, especially call-handling protocols for interfacing between PSTN and IP networks, continue to divert voice traffic onto what were formerly thought of as data-only networks.
RHK estimates that data traffic on North American networks is growing at 80-100% a year and it expects the market for core switches and routers to reach $5,5 billion by 2003. Digital subscriber line (DSL) and cable modems are being deployed in volume in the US, with Europe perhaps six months behind, and falling bandwidth prices are encouraging enterprises to buy more DS3 and OC-3 services. The combined effect is to drive up capacity demands in core switches and routers.
But while many start-up companies are rushing to satisfy demand for ever faster routers - and attracting multibillion-dollar stock valuations in the process - the established vendors are presented with a conundrum: how to help their carrier customers migrate their voice traffic on to data networks while protecting their existing product lines and carriers' legacy installations. Carriers contemplating the move to Packet over Sonet/SDH (POS) must not only meet capacity demands but also provide the levels of reliability associated with TDM and ATM - guarantees that end users expect to see reflected in their service level agreements.
Among the advantages offered by POS are that it eliminates the need to encapsulate packets within ATM cells, thereby simplifying network management and increasing efficiency, for instance by avoiding the 'cell tax' associated with ATM's 53-byte cell and 5-byte header. Instead, advanced hardware routing and buffering directly maps IP packet streams into Sonet/SDH frames with connections based on point-to-point link protocols such as PPP.
However, the sheer cost of such a radical rethink of network architecture makes it impractical except where huge volumes of pure IP traffic are involved.
Instead, emerging, real-world solutions take the best of ATM's mastery over quality control and prioritising of traffic and combine it with the efficiency and ubiquity of IP. The aim of two new protocols now winning support from vendors and carriers is to provide efficient traffic engineering by retuning control to network managers over how their traffic is prioritised and to turn IP into a viable transport mechanism for carrier-grade voice and other realtime services.
Multi-Protocol Label Switching (MPLS) is still under development at the Internet Engineering Task Force, but has received an enthusiastic welcome this year and is already beginning to be deployed in live networks.
MPLS evolved from competing approaches to improving the efficiency of routing IP packets over core networks. Over a period of several years, the IETF assessed 'IP switching' technology proposed by Ipsilon Networks (now owned by Nokia) and IBM's ARIS. But in the end, MPLS most closely resembles the 'tag switching' technology developed by Cisco Systems.
The aim is to build into IP an awareness of traffic load issues, for instance by setting up coarse-grained paths according to traffic engineering requirements and then using fine-grained forwarding to avoid congested routes.
MPLS Label Switching Routers (LSR) deployed at the network edge add a label to each IP packet as it enters the MPLS network. This label contains specific information about the packet's destination, a predetermined path for arriving there, and an indication of service class. Each of the MPLS routers the packet passes through routes according to the defined path - so avoiding the need for an intermediate look-up process while each router calculates the next hop. The LSR, which can be either an ATM switch or router, can also add a service level flag into the IP header. As the packet leaves the MPLS network the extra bits of information are stripped off.
The effect is to separate the overall routing function from each router's forwarding function - allowing the router to be more responsive to network congestion - and to allow network managers to define explicit paths for a particular type of traffic.
MPLS is also designed to protect carriers' existing investment in ATM by controlling how layer 3 IP traffic is mapped onto the layer 2 transport layer. A virtual circuit, in ATM terms, becomes a Label Switched Path (LSP) in MPLS terms. And service class information from the MPLS label can be translated according to ATM quality of service classes: constant bit rate, variable bit rate and so on.
A second, and potentially complementary technology, is Diffserv, also developed under the auspices of the IETF. Diffserv also adds information pertaining to each packet, using 'traffic conditioners' at the edge of the network. But Diffserv (for 'differentiated services') concerns itself purely with Layer 3 and on providing a scaleable approach to QoS for IP.
Diffserv may end up sitting between the enterprise and the carrier network, while MPLS, with its focus on routing efficiency and ATM integration, can be deployed in the core. From there MPLS can enhance Diffserv capabilities, taking QoS data from each packet's 'DS byte' - a modified type of service byte - and translating it into an MPLS label.
Some vendors - including Anritsu with its Multiflow 5000 series multilayer switches - are offering Diffserv features in their products, giving carriers the option of packaging and billing with respect to service classes, particularly for VPNs where granular control over quality would be a welcome feature.
Vendor strategies
Many top ATM switch and router vendors have already declared their support for MPLS.
Ascend Communications said in February 1999 it would release MPLS modules for its CBX 500 switch line. Ascend, part of Lucent Technologies since a $24 billion buy-out last June, is the leader in the overall ATM switch market. Based on 1998 port sales, Probe Research puts Ascend's market share in the US at nearly 38% while Dataquest puts the figure at a more sober 23,2%. RHK reckons Ascend had around 33% in 1998 ahead of Cisco, which increased its market share to 24% by overtaking Newbridge. Fore Systems, NEC, Nortel Networks and others make up the remaining 19%, says RHK.
Fore announced in January 1999 that it would begin building in MPLS support to its AX-4000 switching platform, describing it as "bringing the best of both the IP and ATM worlds to IP service providers".
IBM, which has declared its determination to become a major ATM switch vendor, has committed to MPLS support by early 2000. Ericsson offers MPLS as part of its existing ATM switch platforms. Back in April Ericsson bought Torrent Networking Technologies, maker of gigabit routers that will be packaged as label edge routers within Ericsson's MPLS solution for IP over ATM.
However, MPLS has yet to be tested on large-scale live networks and the persuasiveness of the standard could depend largely on Cisco's success in deploying it. As with world's dominant router vendor, Cisco is using MPLS as a way in to a new market by building IP routing into its carrier-class ATM switches.
Cisco revealed last April that it would deliver its hybrid IP and ATM switches to AT&T Corp - the first US carrier to deploy equipment using the MPLS standard. AT&T will use the switches to support its IP-enable Frame Relay service, with MPLS providing advanced QoS options.
The latest network operator to declare its MPLS strategy is backbone ISP, UUNet. Earlier last year, UUNet lit an OC-48 link between New York, Washington and Atlanta, with plans to roll out similar links coast-to-coast later.
UUNet is looking ahead for a pain-free upgrade path to OC-192. As speeds increase it becomes impractical to continue to interface between ATM switches and routers - an important factor encouraging carriers and ISPs to make the leap to IP.
UUNet Chief Scientist, Michael O'Dell confirmed that with MPLS deployed, the ISP believes it can start to replace its ATM switches, mainly ASX 1000s provided by Fore. Suggesting it will move towards a wholesale move from IP over ATM to IP over MPLS, UUNet said it would run the ATM and IP networks in parallel until the new technology has proved itself.
UUNet's upgrade ambitions also apply to its European networks, which will be based on Juniper Network's M40 Internet Backbone router.
Among the gigabit router start-ups, Juniper, of Mountain View, California, has attracted a good deal of attention. Formed in August 1997 by former Cisco Systems engineers, Juniper's IPO in early 1999 valued the company at $4 billion, when its first quarter turnover was just $10 million. RHK believes the company could seize as much as 20% of the router market from Cisco, based on performance claims for the M40.
The Internet Backbone router handles up to eight OC-48 (2,4 Gbps) ports or 128 OC-3s (155 Mbps) and is considered to have overcome some of the vulnerabilities in standard routing code, BGP. The router can also be fed directly into dense wavelength division multiplexer (DWDM) networks.
However, Juniper stops short of building ATM switching into the M40. Instead the company is betting that other ISPs will follow the example of UUNet. Juniper acknowledges that ATM's hardware-based switching has supported speed requirements until now and ATM Permanent Virtual Circuits (PVC) have played an important role in giving network managers control over traffic loads. But the expense and complexity involved in maintaining both an ATM core and a logical IP overlay and the difficulty of reconciling switch-based traffic engineering with router-based routing, will make the IP-only option more attractive as time goes on, the company says.
Other start-ups, such as Argon Networks, Nexabit Networks (another recent Lucent acquisition) and Pluris, to name a few, are promising to deliver terabit rates. Argon, part of Unisphere Solutions, the integrated voice-data technology company launched by Siemens in March, offers its Gigapacket Node GPN) ATM and IP platform, which is scaleable to 60 Gbps and supports the option to migrate ports from ATM to IP.
Nexabit claims the world's highest capacity router with its 'multi-terabit' NX64000. At 6,4 Tbps - deployable as 192 OC-3s through to 16 OC-192s - the NX64000 is certainly positioned at the outer reaches of router design. Nexabit responds to doubters who question the demand for such capacity by pointing out that DWDM vendors are already talking in terms of OC-768 and OC-3072.
Significant challenges
"Although many believe that network traffic will eventually be consolidated onto a single IP-over-optics infrastructure, many service providers will find that keeping two functionally different backbones - one entirely IP-centric and the other based on ATM - will result in a more efficient network," says Joe Skorupa Director of RHK's switch and routing service.
But service providers contemplating multiservice networks face "significant challenges," says Skorupa. Existing voice/TDM networks "have set very high customer expectations" for robustness and quality. IP, ATM and hybrid networks must match these expectations while delivering step improvements in scalability, capacity and cost reduction.
© Technews Publishing (Pty) Ltd | All Rights Reserved
printer friendly version