Hot Spot - Can MPLS Span Access to the Core?

Many know that it is a way to make IP networks behave more like switched networks, with the ability to provision and guarantee bandwidth across a network. Others look at MPLS as if it is the Holy Grail of voice quality in IP networks, the hope for making IP networks behave like circuit-switched networks where quality counts.

Lately, it also has been proposed as the basis of service provisioning in core and metro optical networks by a slew of vendors, including many year 2000 startups, such as Alidian Networks Inc. (www.alidian.com), Appian Communications Inc. (www.appiancom.com), Astral Point Communications Inc. (www.astralpoint.com), Chromatis Networks Inc. (www.chromatis.com), Sirocco Systems Inc. (www.siroccosystems.com), Sorrento Networks Inc. (www.sorrentonet.com) and Tenor Networks Inc. (www.tenornetworks.com).

MPLS originally comes not from IP networking, but from ATM, where it was enabling ATM switches to increase in capacity faster than routers. Study groups now have been established in the Internet Engineering Task Force (www.ietf.org) to develop the technique into Internet standards that will do tasks from traffic engineering to QoS for voice.

What is MPLS?

MPLS adds a "label" to the headers that are on every IP packet, but the key is that labels are shorter than headers--about 20 bits. "IP headers were usually longer and had to be matched to a big [routing] table" of IP addresses, says David Drury, vice president of technology strategy at Marconi plc. (www.marconi.com), and chairman and president of the MPLS Forum (www.mplsforum.org), an industry group promoting MPLS. Labels can be switched much faster than full headers because there is less for the router to process.

Also, label lists are shorter. The router says to itself, "I have the label, just check the label list," and it puts a new label on it and sends it out a different port.

Also, labels specify full paths from the source to the destination, paths that have been set up ahead of time by MPLS signaling. Rather than trying to figure out the hop-by-hop route that is best at any given moment for the packets, a label just directs a packet on an end-to-end route that is already specified. And the route is specified right from the "ingress router," the router where the packets enter the network.

"If you create a protocol that sets up paths, now you have a means to do traffic engineering," says Drury. "Now you can take into account the capacities of all the links between A and B, the capacity of the flow I need to place, and set up labels that define the end-to-end path, and that is a label-switched path [LSP]."

For the first time, labels give a way to control the path traffic takes through an IP network. "Traffic engineering allows you to spread the load out, but to do that you need explicit routing and the ability to dial up a connection," Drury says. "This is essentially introducing the whole connection paradigm into IP, which has a connectionless paradigm."

The way those paths get set up in the first place is one of the most important capabilities of MPLS. The ingress router uses MPLS to get information about the network, such as its topology, how big the links are and which ones are in use. It also gathers information about the flow it is about to send, such as its size and destination. It takes all this information and uses a formula, called a path-computation algorithm, to calculate the best path.

"Because with labels the only real routing is done at the edge, it makes it easier to scale the network," says Tony Gayle, director of product management, MPLS for Lucent Technologies Inc. (www.lucent.com). "A path is set up across the network and you don't have to look at every packet. This will be very important for service providers in the future."

After the path is chosen, there has to be a way to tell all the other devices on the network what the path is and the size of the flow. This is MPLS signaling, and it is here that there is some disagreement among vendors. Two signaling systems are in use: An extension of resource reservation protocol (RSVP) signaling called RSVP-TE (traffic engineering) and constraint-based label distribution protocol (CRLDP). While RSVP-TE now is more widely used, there is no clear consensus.

Taking the middle road are Marconi and Lucent. "We are using MPLS-TE, working with Marconi, in our ATM network," says Gayle, but the company also has demonstrated CRLDP, which it uses in its ATM products.

Another way to use labels is to stack them. The first label may denote the service provider delivering a particular flow, but the second label may be used to denote individual customers of that service provider. Stacked labels help differentiate and separate traffic.

Another advantage of using MPLS signaling is that it allows a path to be re-established automatically if there is a break. The signaling is merely resent and the path is restored.

Optical Routing

The simplicity and universality of MPLS signaling lately has become attractive to optical vendors, particularly those that create DWDM products for core networks. They are proposing to use MPLS to switch not flows of packets, but entire waves or lambdas.

The impetus is the rapidly changing environment for optical networks. "SONET and SDH [synchronous digital hierarchy] essentially statically provision a network. When you created a SONET trail from A to B, it stayed up forever," Drury says. "That was OK with voice where things don't change much, but in the Internet customers need to provision rapidly, immediately, not in weeks."

A revolutionary proposal to the IETF about 18 months ago proposed MPLS as the solution. The simple concept, however, is not so simple to execute. "There are nontrivial issues because there are different constraints about routing waves rather than flows," says Drury. "The granularity is much higher, and you can turn on and off only lambdas. You can't change a wavelength's [contents], and you have to find the same wave all through network."

Chart:MULTIPROTOCOL LABEL SWITCHING ARCHITECTURE OVERVIEW

Still, the idea has attracted enough attention that a new IETF group has been set up to develop the protocol. Initially, it will be MPLS-like, but with specific features for optical networks. The expectation is that eventually the two protocols will merge.

MPLS QoS?

The burning question remaining for vendors and service providers alike is, will MPLS solve the overriding issue of voice quality in IP networks? "You can now do pseudo-QoS," says Drury. "When you set up a path, you can reserve 10 megs, for example. Now, having got a connection, you can start asking for the characteristics of that connection, and this is where we start moving up to classes of service in IP. This is like implementing some sort of connection paradigms in IP."

But to guarantee that a flow will get those 10mbps, MPLS signaling has to police all the other flows to be sure they don't do anything that would interfere with the availability of the 10 megs. So MPLS will have to do scheduling and manipulation as services move beyond best effort.

Despite the obvious benefits of MPLS, it is not a panacea, emphasized Tom Jenkins, principal analyst, TeleChoice Inc. (www.telechoice.com). "If I have MPLS and have an over-utilized network, I will still not have quality. MPLS is important, but it is just one piece of the puzzle. If I have a poorly managed network with it, it could be better than without it, but you would still have poor voice communications." He adds, "There is no current 'be-all' out there that combines software and hardware management."

Another barrier to MPLS is that vendors are doing their own implementations while waiting for a final standard, and carriers have not chosen to push for interoperability, using instead islands of single-vendor MPLS. "The ones who really could force the standard are not in a position to take advantage of the technology if it were to happen," Jenkins says, "They have the most to lose because they have existing infrastructure and prefer to slow migration."

Those who really want standardized MPLS are emerging providers who don't have the clout to demand a standard.

For MPLS equipment today, there are interoperability labs under way at the University of New Hampshire and George Mason University in Fairfax, Va. They are starting with simple operations, such as exchanging labels, then will move up in complexity.

Most deployments are single vendor, usually within ISPs and used for "fairly gross traffic engineering, but not turning on all the signaling or features," says Drury. "But over the next six to nine months I expect to see it ramp up quickly. This is in the core of large IP networks, which is the initial target because it is where the problems are greatest."

As many as 30 small companies also are preparing all-MPLS edge switches, he says, that will be deployed beginning in the second half of 2000.

A true optical standard is as much as a year away, but interim implementations have already begun, with vendors designing their technology to change as the standard matures.

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