Is there a 'Net QoS standards standoff?

By Jim Duffy
Network World, 11/02/98

Atlanta - Are two Internet Engineering Task Force working groups advancing at cross purposes with regard to quality of service (QoS)?

A debate has arisen about whether the IETF's Multiprotocol Label Switching (MPLS) and Differentiated Services (Diff-Serv) efforts are competing to accomplish the same goal of equipping the Internet to deliver QoS.

Some say competition between the working groups threatens to deprive users of Internet QoS; others say the efforts are mutually exclusive and shrug off the competition as standard operating procedure within the IETF.

"There's a certain creative tension between the groups," says Paul Doolan, chief technical officer at Ennovate Networks and a co-author of the MPLS specification.

"I think creative tension might be a more forgiving way of putting it," says Tom Nolle, president of telecommunication consultancy CIMI Corp. of Voorhees, N.J. "The current standards polarization . . . absolutely could poison almost every advanced application of the Internet."

MPLS is designed to help scale the Internet by introducing circuit-switching techniques to a packet-switched, connectionless environment. MPLS adds a "label" to IP packets that steers that traffic through the Internet over predefined routes.

MPLS helps scale the Internet by alleviating the need for each router and switch in a packet's path to perform address lookups. MPLS also provides more deterministic, or predictable, performance, which is essential for guaranteeing QoS.

Diff-Serv is intended to deliver different levels of QoS to different applications. Diff-Serv proposes using the type of service (ToS) bits in an IP header to assign QoS classifications to different traffic based on service-level agreements hammered out between users and service providers.

Diff-Serv and MPLS are considered packet tagging techniques for delivering QoS. But Diff-Serv works at Layer 3 of the Open Systems Interconnection model, while MPLS works at Layer 2, meaning the efforts should be mutually exclusive of each other. MPLS can work fine with or without Diff-Serv, and vice versa.

"The basic point is that MPLS and Diff-Serv are not trying to do the same thing," says Scott Bradner, a consultant with Harvard University's University Information Systems. "MPLS does intend to deal with a class of QoS issues, but that has not been its main thrust, which is traffic engineering. Diff-Serv's main function is to look at QoS issues and to create building blocks for QoS services. I do not see why we cannot figure out a way that Diff-Serv networks can make use of MPLS underlying networks."

MPLS devices might assign switching labels after reading Diff-Serv instructions in IP ToS headers. The issue, according to MPLS co-chair George Swallow, is whether Diff-Serv can use three bits in the MPLS label byte for this.

"The question for the current MPLS encapsulation is, 'Is eight bits enough?' " Swallow says. "If it is, then the [Diff-Serv markings] can be encoded in the three available bits. If not, we'll need a new encapsulation."

It may be this nuance that's sparking the firm stances some vendors are taking on MPLS and Diff-Serv. Ennovate's Doolan characterizes Diff-Serv as a "grand scheme" vis a vis MPLS, which he says is a "pragmatic attempt to build useful networks."

Xedia, a maker of IP switches for ISPs, is squarely in the Diff-Serv camp, says company president Ashley Stephenson.

"MPLS is vendor-driven," Stephenson says. "In many ways, I think MPLS grew from the leftover IP switching, from Cisco's Tag Switching and from ATM being threatened by packet-over-fiber. So there's lots of vendor reasons why a lot of very smart people have got a lot invested in something like MPLS."

"MPLS and Diff-Serv compete for the same market," says Shai Herzog, founder of QoS company IP Highway, and co-author of Diff-Serv and the Resource Reservation Protocol. "That market is the market that talks about end-to-end QoS delivery. It's not really a technology issue, it's a marketing issue."