Developing Efficient Managed IP Networks

People in this industry spend a lot of time talking about how the Internet is a best-effort network. That’s often the starting point that leads to a discussion about how their network is a much more controlled environment that delivers quality of service for such time-sensitive traffic as VoIP. But what does it really take to create and operate a managed IP network? And what still remains to be done to most efficiently engineer and deliver on the promise of such networks?

“Probably the most common thing people refer to when they talk about a managed network versus a non-managed network is kind of the contrast between a service like Cbeyond, where we have control of the Layer 3/routing network versus a service like Vonage or Packet8 where they’re relying on existing broadband to deliver a Layer 3 service like voice over IP. Probably the main difference is quality of service, says Chris Gatch, CTO at managed services provider Cbeyond Inc.

“At the core, there’s a good bit of bandwidth available, but when you get down to the last mile and you’re dealing with a T1 — 1.5megabits a second, or something in that range — it’s very easy to have contention for bandwidth between data applications and voice applications,” Gatch says. “A single PC downloading a file can consume a single T1 easily. So if you don’t have prioritization of your voice traffic over your data traffic, then it leads to call quality issues, calls dropping, things like that.”

Delivering QoS means controlling the queuing both from a customer’s location and on the downstream path, he says. “Ultimately, you have to be in the routing path to decide which packets you’re going to let go into the priority queue versus which ones are going to be handled by the default class,” Gatch adds.

Deciding which traffic goes into which of the two queues is a very simple proposition for Cbeyond, which runs an all-IP, Cisco-powered network. “Basically, all of our outbound interfaces on any low-bandwidth interface — and for us that’s DS3 or less — we run priority queuing. On our large, core interfaces, we don’t have to run queuing; it provides diminishing returns in the core because it’s much easier to run an IP network with excess bandwidth than it is to try to manage it tightly with QoS.”

Meanwhile, at Global Crossing Ltd., VoIP gets top priority on the company’s private MPLS core network, says Al DiGabriele, the company’s vice president of network applications. “We do SLAs on our backbone,” he says. “We also have multiple access options for our VoIP service.”

In order to bring QoS to voice, policy-based multimedia call admission control is necessary, says Adam Uzelac, director of converged services architecture at Global Crossing. Using QoS mechanisms within IP only does half the job, he says. To deliver true QoS, a network operator needs to start crawling up the OSI ladder to see what’s really going on, he continues.

Today, the common practice is to parse out the user’s access bandwidth between various types of services. It’s kind of like the ATM-based method of nailing up virtual circuits. The goal is to move toward more dynamic bandwidth allocation within a given connection.

That will enable a user who subscribes to a 2mbps service to use the full capacity of that link for whatever he wants, says Uzelac. So the ideal is to have an intelligent device at the customer premises or within the network to allow the customer to use the whole of his connection for any application on an as-needed basis, but not to exceed that bandwidth, he says.

Dan Bantukul, director of product marketing at signaling gear supplier Tekelec, says policy enables that by managing the call admission control as well as working hand-in-hand with the IP layer to allot the necessary bandwidth to a given application so multiple classes of service peacefully can coexist on shared bandwidth.

Global Crossing has policy-based call admission control on its road map, probably in the latter part of 2008, and now is supplying requirements to vendors for IADs to be used for converged customers and for edge devices that do policing of multimedia networks, says Uzelac. He adds that this new generation of IADs will “leapfrog” the current generation due to significant jumps forward in terms of memory to handle new software loads and identify sessions, and these edge devices will be the future for session border controller devices.

Right now, however, many service providers that prioritize VoIP on their networks employ DiffServ to do so, says Andy Randall, vice president of marketing at softswitch supplier MetaSwitch. “That seems to get pretty good quality,” he says. “The issue is designing a network so every hop respects that prioritization.”

The MetaSwitch gear supports QoS in a number of ways including by supporting DiffServ, he says. And, he continues, MetaSwitch is adding functionality not typically found in softswitches. That will include the ability to gather and aggregate statistics from all network elements involved in a call so the service provider can see in real time what’s happening in the network.

But controlling QoS within an individual network is just part of the solution, adds Mark Neider, senior director of product marketing at session border controller vendor NexTone Communications. Network operators can’t deliver a complete QoS solution without monitoring and managing the traffic that passes their interconnection thresholds to and from other carriers, he says. If traffic is not at the quality a service provider requires on its network, Neider explains, the NexTone session manager dynamically can reroute that traffic to another of the network operator’s service provider partners’ networks.

Cbeyond Inc. www.cbeyond.net

Global Crossing Ltd. www.globalcrossing.com

MetaSwitch www.metaswitch.com

NexTone www.nextone.com

Tekelec www.tekelec.com

Share this article: Email, Slashdot, Digg, Del.icio.us, Yahoo!MyWeb, Windows Live Favorites, Furl
Add this article feed to: RSS, My Yahoo, Newsgator, Bloglines