The Whens and Whys of IP-Based DSLAMs

Posted 04/01/2001

The Whens and Whys of IP-Based DSLAMs
By Paula Bernier

DSLAMs are evolving from relatively dumb DSL traffic aggregators that use ATM for transport into more intelligent devices that often employ IP transport in addition to the currently prevalent ATM.

While no one doubts that ATM is now--and for some time will continue to be--the backbone transport choice over which DSL and many other access services will travel, some DSLAM vendors are adding new IP transport functionality to their wares.

ATM was the clear choice initially for ILECs because they had ATM backbones in place for existing services such as frame relay. Plus, ATM has QoS capabilities built in, whereas IP was created as a best-effort protocol. Only in the past few years have companies and industry groups begun to address differentiated services over IP networks.

And, most agree, lack of standards for IP QoS and incumbent carriers' comfort level with ATM will help ensure the status quo of DSL over ATM for the foreseeable future--at least for the big carriers. At the same time, some vendors are saying that DSL is outgrowing ATM. They are introducing new IP-centric features to their DSL muxes and switches to enable carriers to realize new savings by taking advantage of some of the efficiencies IP transport can offer.

"There is a large embedded base of ATM transport out there now. For ILECs, it will be an ATM solution for the next few years. But all are evaluating IP solutions too," says David Williams, market manager for ADTRAN Inc. (www.adtran.com), which, during the first half of this year, plans to announce a DSLAM that can be upgraded with a mux card to native IP.

"If you look at why CLECs are suffering, it's because the cost of operations of DSL is high, and a big part of that cost is the provisioning costs," says Enzo Signore, director of marketing for Cisco System Inc.'s (www.cisco.com) DSL business unit. "All ATM-based equipment costs and provisioning of ATM PVCs [permanent virtual circuits] is expensive. If you have IP routing, you don't have that problem because the protocol itself takes care of provisioning."

DSL providers need to connect consumers to their ISPs. But those Internet providers have hundreds of users going to a single destination--the ISP. Rather than provisioning a separate virtual circuit, or VC, for every service of every different consumer, all that traffic destined for the ISP can be bundled on a single IP connection, he says.

"ATM makes you associate each user to a VC," Signore says. "With the IP network, you don't have to use a tunneling protocol; instead you use an MPLS solution. Provisioning costs have to do with the number of tunnels."

The Cisco 6000 IP DSL Switch, which is considered a next-generation DSLAM, has full IP routing capability so the carrier can choose to route its DSL services over an ATM or IP backbone, he says. The vendor announced MPLS VPN capability on the product last year. In the future, the company expects to announce the availability of additional software to enable public network operators to carry additional IP protocols directly on the DSLAM, eliminating the need for separate tunnel/VC aggregation products like those from Shasta Networks Inc. (now owned by Nortel Networks Corp., www.nortelnetworks.com) and Redback Networks Inc. (www.redback.com), Signore says. In its second software release, available this year, Cisco expects to add point-to-point protocol (PPP) termination and Layer 2 tunneling protocol (L2TP) features to its product.

Key Elements in xDSL Service Rollout Process

Copper Mountain Networks Inc. (www.coppermountain.com) also already offers an IP transport option in its DSLAM, and has also announced its intention to add functionality on its product for PPP aggregation into L2TP, enabling the box to take in up to 200 PPP dial-up sessions and tunnel them using L2TP over a single VC, according to Bryan Long, vice president of marketing at Copper Mountain. The company plans to make specific announcements about the availability of the new feature later this year.

According to Long, the scale of DSL deployment is outgrowing the ability of the ATM backbone--which handles frame relay and other services in addition to DSL--to support it.

"DSL is more of an Internet phenomenon, so the scale is swamping ATM because VCs are taking up the network so quickly," Long says. "Now everybody is using DSL only for Internet access, so there is only one VC per subscriber for the single service of access to the ISP. But as voice is added, another VC will be needed per subscriber."

Carriers will probably continue using VCs even if they use IP for transport, Long explains, but in a different, more efficient way. Today, carriers haul VCs through ATM switches and terminate the circuits on a Redback or other similar box. But next- generation DSLAMs in the future could handle that DSL termination, look at and prioritize the IP packets within, and then aggregate them based on priority before handing that traffic off to the ATM backbone. That would eliminate the need for the extra box and would only require one VC per DSLAM per service, he says (or the carrier could strip off the voice at the DSLAM and hand it off to a voice switch).

"That's what our pitch has been," says Long. "I'm saying we need to displace Redback totally. It could still do things like authentication, user selection, etc. But ILECs are buying banks and banks of Redback stuff just to do VCs."

Once that kind of intelligence hits the DSL concentrator, he adds, carriers might question why they need to put their traffic back on an ATM backbone when they can just use gigabit Ethernet switches.

There is also a well-worn argument that using IP vs. ATM will result in overhead bandwidth savings.

A pure IP system takes up less bandwidth on overhead than does ATM, says Keith Atwell, director of business development for ADTRAN. Also, it doesn't require the protocol conversions and resulting delays at the WAN to LAN interface, says Atwell.

"Also, midyear we'll see IPv6 [IP Version 6], which will have good QoS in the local LAN," Atwell says. That will push more telephony traffic onto the LAN using IP, so carriers as a result might want to move to native IP in the WAN too, he adds.

Fima Vaisman, vice president of marketing at Accelerated Networks Inc. (www.acceleratednetworks.com), says there's no such thing as the IP-based DSLAM. IP always has to run over something, he says. But in any case, he says, the overhead in ATM is significantly lower than that of IP.

ATM cells each consist of 53 bytes, five of which are overhead, the rest are the actual payload. An IP packet, meanwhile, has variable byte headers--it could be 10 or 12 bytes, depending on format. "IP packets also can't get on a network until going on another protocol, which is yet more overhead--at least another five bytes," Vaisman says. "With ATM, less bits and bytes have to be transported left and right."

Whatever the case of the overhead and QoS debates or the benefits IP could offer on the virtual circuit front, people remain convinced that widespread use of IP-based DSLAMs remains at least a few years off.

"We surveyed over 60 service providers on that," says Jennifer Stagnaro, chief marketing officer at CopperCom (www.coppercom.com), which specializes in the VoDSL space. "Those ISPs, RBOCs and CLECs said they would only entrust voice to an ATM-based network at this time. That was three years ago. Still, ATM is the way to go."

Jay Fausch, senior of director of marketing at Alcatel USA (www.alcatel.com), says that while his company expects to add an IP resource card to its DSLAM early this year, the vendor doesn't expect to see a lot of demand for it.

"Service providers deploying in the mass-market model are going to continue to support ATM because of QoS," he says, while IP standards like MPLS and differentiated services (DiffServ) that focus on quality or class of service still haven't been solidified.

"It's a timing issue," adds Tim Waters, vice president of marketing for the broadband access business unit of Nortel Networks. "If you see what's been shipping in 2000 and anticipated for 2001, the vast majority of that will be ATM-based products.

"Look at what's happened in the DSL access space," he adds. "People have gotten a real lesson. They've come to the conclusion that $29.99 is not a long-term business plan. So you need to differentiate service, and ATM QoS allows them to do that."

Nortel is agnostic as to whether the DSLAM on the transport side goes IP or remains ATM, Waters says. The issue as to which way it goes will be when IP "gets hardened" and network engineers see standards in place and are comfortable with the level of reliability it can offer.

Steps in the Deployment Process