ISPs fail to fully follow through on QoS

Getting beyond best-effort service from an ISP remains a challenge, as most service providers still have no plans to offer guaranteed classes of service over their IP networks.

While technologies such as the Internet Engineering Task Force's (IETF) Multi-protocol Label Switching (MPLS) and Differentiated Services (Diff-Serv) are often talked about, they are not yet fully implemented.

Off-course

Many carriers use MPLS to engineer traffic by setting up paths through the net for specifically labeled packets. Diff-Serv also tags packets to set priority levels, but this capability is not widely deployed by ISPs.

So what's the holdup?

"It's still pretty complicated, and many carriers are still waiting for standards that would support QoS over multiple networks," says Maribel Dolinov, an analyst for Forrester Research.

Although MPLS is available, installing it can be cumbersome because nodes throughout the entire network need to be upgraded, Dolinov says.

Service providers that have homogeneous networks may have an easier upgrade, but many aren't finding enough reasons to take on the costs. "Carriers aren't seeing a return on their investment, nor can they figure out how to charge users for such services," Dolinov says.

WorldCom is using MPLS for traffic engineering over its UUNET Internet backbone. "MPLS combined with high speeds ensure you don't overload your link," says Dave McDysan, a WorldCom fellow in charge of traffic engineering. MPLS allows WorldCom to direct traffic, and high bandwidth allows the buffers on the switches to be emptied more often, he says. But it's not a networkwide upgrade that could offer users guaranteed services.

WorldCom already lets users dedicate bandwidth to specific applications or user groups, but is looking at using a version of Diff-Serv to offer users multivendor QoS support.

The IETF is working on a specification for Diff-Serv to let carriers set QoS levels that span multiple ISP networks. While still far off, McDysan says this could be the first incarnation of multiple-network service-level agreements.

Once the IETF defines how traffic should be forwarded using the new specifications, ISPs would still have to negotiate transit agreements and settlements, he says. Adding a new level of enhanced services to this arrangement will take more time, especially because ISPs are having difficulty determining how to bill their own customers for added QoS, Dolinov points out.

AT&T is offering an Internet access service that uses Diff-Serv to let users prioritize traffic from their router to AT&T's edge router. "We did this because the access link is where users see the biggest problems, especially in areas where it's difficult to get large circuits," says Rose Klimovich, director of IP services at AT&T.

Voice and other real-time applications can get top priority, while other traffic such as HTTP can be given lower priorities. But this class-of-service feature only goes as far as AT&T's edge router. Once a user's traffic hits the Internet, it's back to best-effort service.

AT&T plans to use MPLS but hasn't started deploying it. "MPLS will be deployed within our network for traffic management later this year, which is what MPLS is best at," Klimovich says. "We may use it to build VPN services down the road."

Cable & Wireless will likely use MPLS to offer users a VPN service, but it has to ditch its ATM core and deploy IP MPLS gear, says Dave Garbin, senior director of network architecture at the service provider.

Cable & Wireless, like many of the large national ISPs, originally deployed ATM switches to handle higher traffic volumes at the core of its network. But ATM eats up more bandwidth than IP and doesn't scale well, Garbin says.

To better direct traffic at the core, Cable & Wireless has deployed Juniper switches in its network using MPLS. The company plans to have MPLS fully deployed by the end of next March.

Sprint is one of the only large ISPs that is not deploying MPLS. It believes MPLS is unnecessary.

"We continue to maintain that our basic engineering . . . is still the best way to go," says Chris Clark, vice president of product management for Sprint E-Solutions at Sprint. "What you'll see today and in the future is IP over glass. We don't need another system like MPLS that adds to overhead."

Sprint says its network will be IP over fiber optics no later than next January.

Sprint is looking at technologies such as Diff-Serv for QoS, Clark says.

In addition to moving away from legacy technologies such as ATM in the core, Sprint is leaning on a tried-and-true traffic engineering technique to keep traffic loss and jitter as low as possible. When a path on Sprint's Internet backbone reaches 50% utilization, Sprint adds more bandwidth and resources to that path, Clark says. In essence, Sprint is throwing more bandwidth at its problem areas and believes management and monitoring is the best way to offer users stable QoS.

"That's a good old-school approach," Dolinov says. "The fact is that in the U.S. there is a lot of dark fiber out there." As long as this is true, bandwidth is still one way to avoid congestion and provide QoS.

ISPs fail to fully follow through on QoS

Related Links