Focus: Ethernet MANs
Rolling out Ethernet MANs

Ethernet is an old technology and the carriers may be hesitant to adopt it. But there are some key benefits of implementing an Ethernet service over a Metro network. by Ong Boon Kiat

It is ironic that the almost ancient Ethernet technology is now seen as the next big step for the carrier metro networks. But it is hardly surprising. Given the ubiquity of Ethernet in corporate LANs, it is natural that Ethernet will also make the best and most cost-effective WAN technology. At the very least, it is more elegant than the various ways of retrofitting telecom lines and cables for data transfer over long distances.

So why is Ethernet not yet a key ingredient in carrier networks today? There are two main reasons. Firstly, most carriers are tenacious preservers of legacy investments—and rolling out Ethernet-based Metro Area Networks (MANs) means more expense will be incurred in laying new fiber optic cables and replacing modems with IP-switches. Secondly, the demand for Ethernet over service provider networks has not been exactly ravenous. But it may soon be.

Several prominent telco experts who spoke during keynotes at the recent CommunicAsia event in Singapore (June 2002) were strong in their support for Ethernet-based MANs, which they say will form a crucial part of carriers' next-generation networks. One keynote speaker, Anand Parikh, founder of Appian Communications, presented a rather convincing cost-benefit analysis for Ethernet MANs. He showed that the technology will not only save money for enterprises, it will also dramatically improve carriers' service provisioning, billing, and voice-integration capabilities.

In North America, spending on MANs will increase five-fold over the next four years, from USD 420 million to USD 2.7 billion in 2006, according to Infonetics Research. It reported that most businesses with LANs use dial-up, DSL or T1 connections to connect to their WANs or the Internet.

In this respect, Asia does not lag behind. According to market researcher IDC, the number of metro Ethernet subscribers in the Asia-Pacific (excluding Japan) will grow from 0.28 million in 2001 to 9.29 million in 2006, representing a CAGR of 110 percent. The revenue proposition looks even better. IDC predicts that the same market will generate USD 1.5 billion in revenues by the end of this year, from USD 395.34 million in 2001. And it will grow at a CAGR of 118 percent to USD 19.44 billion, from 2001 to 2006.

Why Ethernet?
So what are the benefits of implementing Ethernet service over metro networks? Michael Kelly, CTO of XA TMI, offers two. First is flexibility in bandwidth allocation. For example, enterprises can obtain multiples of 1 Mbps bandwidth pipes instead of a large jump from 2 Mbps to 45 Mbps when more than 2 Mbps is required. In addition, Ethernet is also well adapted to bursty traffic, unlike rigid leased lines, which need to be provisioned first.

Secondly, enterprises can save on hardware since Ethernet-based customer premise equipment (CPE) are cheaper than telco interfaces, he says. There will also be savings from hardware management, as Ethernet is more widely known and supported.

Another Ethernet strength is its flexibility. For instance, enterprises can easily scale up or down when they upgrade, downgrade and even relocate their Ethernet services. They can also deploy multiple services over a single Ethernet interface. This benefits service providers, since only one link is required to connect to all their clients. This level of simplicity ultimately means less roll-out costs for service providers.

Yipes, a US-based Ethernet MAN provider, published a white paper that detailed the merits of having an Ethernet 'dialtone' in enterprises. It opined that having Ethernet at a WAN level lets network administrators attain virtually all design objectives of an enterprise network. These objectives include the ease of connectivity, flexibility to scale and change network design, end-to-end quality of service, high operational redundancy, low administrative support and low capital expense.

A few flavors
The fastest Ethernet MANs today are based on fiber optic links, offering bandwidth up to 1 Gbps. They can be offered as OSI layer 2 and 3 services. At layer 2, which is the data link layer, enterprises will be able to control the network connectivity of the service. At layer 3, which is the network layer, the service provider takes care of the network connectivity between branches. Layer 3 Ethernet MAN services must therefore be well-managed by the service provider.

Regional data and voice service provider XA TMI, has Ethernet offerings in several Asian cities that provide bandwidth in multiples of 1 Mbps using its customers' existing layer 2 technologies. XA is also looking at extending its layer 2 Ethernet service across its regional backbone network. Singapore's StarHub also offers a fiber-based Ethernet service called IP.Q, which is designed to support IP protocols at both OSI layers at up to 1 Gbps. It is linked to 600 buildings in Singapore's central business district and other industrial parks.

StarHub currently offers only managed Ethernet services, but says that it will soon sell layer 2 offerings. Another Singapore-based regional provider, SingTel, currently offers a layer 2 fiber-based Ethernet service, called Gigawave.

So which Ethernet service class is superior? There is no clear answer here, says Andrew Grenville, vice-president for product marketing and management, business markets, StarHub. He says that depending on the types of networks and the applications that customers have, both modes of Ethernet have their merits. Just as there are times when a provider-managed service is desirable for ease-of-use, there will be environments where enterprises need complete control.

"Layer 3 Ethernet service can be very useful for customers who are building multi-site VPNs and would like to vary QoS based on different requirement levels," he says.

On the other hand, he added, layer 2 provisioning lets customers design their own traffic prioritization, VPN routing and applications such as online storage and recovery where Ethernet class speeds (10 Mbps or higher) are necessary.

For service providers however, it is important that they support both layer 2 and 3 network protocols, says Grenville. This will let them better administer SLA commitments and provide more transparent connectivity to customers. Offering both modes of Ethernet access also gives customers a choice.

A key feature of IP.Q—and increasingly, for most new IP data services like SingTel's Mega@POP—is the use of MultiProtocol Label Switching (MPLS) at the network backbone. MPLS is a way of tagging traffic as it enters the edge of a WAN, so that MPLS-enabled switches and routers can ensure that each data stream flows in a preferable network path, thus ensuring QoS. The best part about MPLS is that it is transparent—tagging can be applied at the ingress and stripped at the egress, without involving any application at all.

Using MPLS, service providers like StarHub can extend their MAN service beyond local fiber hubs, since an MPLS core can be used to link different MANs using a shared backbone. A MPLS-based IP core can also be used to aggregate ATM networks, although whether end-to-end quality of service can be maintained in a hybrid ATM/MPLS/IP network is still open to argument.

Don't write off legacy
On the horizon, the rise of Ethernet MANs does not yet spell the death of entrenched telephony-based access like ISDN, DSL, leased lines, or even ATM services. Since telecom operators are well-known for making the most out of legacy investments, expect a long fall-off period, if any, for these offerings.

"Ethernet will be popular as one of the important data interfaces for the enterprises, but other interfaces like ATM and leased lines will retain their dominant position," says Goh Boon Huat, SingTel's deputy director, IP Services.

Instead of supplanting leased circuits, Goh sees Ethernet MANs as filling a growing market segment which demand transparent LAN access, and are located in fiber-concentrated business districts. For example, with optical Ethernet, it is easy to offer 2,3,4,5 Mbps services, whereas traditional services would see a big jump in bandwidth from 2 Mbps to 34 Mbps. In this respect, it will be able to fill a gap in high speed data services.

But even optical Ethernet, he says, should be viewed as merely a more flexible interface, in that the bandwidth is more flexible. "How useful the customer will find this is still dependent on what this Ethernet is connected to. For instance, are we talking about Internet access, VPN access or transparent LAN?"

His advice to MIS seeking WAN services is to not place undue emphasis on the network interface types, but on how the service will be applied. "Remember that even Ethernet cannot exist by itself as a service, and has to be part of a managed offering," he says. In the end, enterprises must understand what the real value proposition of a service is before taking the plunge. This means that careful analysis of traffic types and patterns will be required before scoping out a new service requirement.

In short, Goh is saying is that MIS should take the Ethernet hype with a dose of reality. Indeed, experts have questioned the reliability of Ethernet as a mission-critical transport layer. For example, some experts have pointed out that Ethernet restoration takes a few seconds compared to SONET's 50ms, which is a crippling blow to the former's reliability claims. This and SONET's inherently more secure logical paths means that SONET still retains an edge as a WAN transport—at least in the near future. In addition, SONET's time-division multiplexing (TDM) capabilities means that it is still the undisputed leader in carrying voice traffic.

Future WANs
Looking forward, StarHub's Grenville sees an inevitable escalation of bandwidth and new value-added services. He says that StarHub is already planning to extend IP.Q to 10 Gbps.

In particular, the need for enterprises to have data residing outside its own LAN for disaster recovery purposes, but still managing that data as part of its LAN environment, will drive IP-WANs and Ethernet MANs, he says. As for service types, hot trends will include global-managed services, VoIP, VPN, multicast and video-streaming applications.

In the longer term, he expects Ethernet MANs to converge with mobile networks such as GPRS and 3G to extend the reach of corporate data networks.

As for optical Ethernet, he believes that it will eventually spread into the satellite cities around main fiber-laden hubs. When that happens, cross-regional Ethernet solutions will become viable. "Of course, traditional telecom service providers will try to use the existing telecom infrastructure to provide the optical Ethernet service, while new startup telecom service providers will build a new MPLS platform or switch platform to provide the service," he says.

XA TMI's Kelly expects lots of infrastructural upgrading work for carriers in the next few years as they retool for Ethernet and IP. For one, carriers are likely to focus on developing switching capabilities that can handle point-to-multipoint and rapid failure detection so that their IP services can become more "carrier class", he says.

But don't expect to see the zealousness in spending that typified carriers five years ago. "Today's telecom environment does not favor providers installing a lot of new equipment for new services whose take-up may be slower than anticipated," he says.

Instead, most will look to retrofit existing equipment to offer Ethernet services. He noted that some equipment vendors are already adding Ethernet features in their router and switch upgrades, which means service providers and telcos simply need to ramp up the capacity between the switches and routers to roll out high-bandwidth Ethernet services.

In the longer term, some carriers will begin to use Ethernet MAN for voice services as VoIP at the PBX becomes more widespread, he says. But he does not think that Ethernet everywhere will happen soon.

Says Kelly, "Other than multiple 1 Mbps service offerings, Ethernet over international lines will take longer because cost is still high on many routes."