QoS with More Than One MPLS Provider is Tough

One of the biggest selling points of carrier-provided MPLS is integrated QoS at the edge and within the cloud. While the purest QoS is on dedicated leased-lines, that technology fails to scale on global WANs that would require many long-distance links. Cloud technologies remove the distance cost associated with leased-lines. Frame-relay and ATM were two of the earliest cloud technologies that eliminated the distance issue for global networks, but their QoS technology was limited, particularly with frame-relay which provided the bulk of low-bandwidth links.

Enter MPLS with better routing (any-to-any on 2547bis VPNs) and integrated IP QoS. Most MPLS carriers offer some sort of QoS. I have not found one yet that does not offer QoS with Layer-3 VPN (2547bis) service. Most MPLS QoS for customers is based on IP precedence with some carriers providing enhanced traffic differentiation based Assured Forwarding (AF) classes. All carriers provide a separate, top queue for VoIP. This is all very good. With QoS customers can prioritize critical traffic over bulk data transfers, Internet, backups, and garbage (scavenger) traffic.

The problem is none of the carriers have the same QoS model. Verizon Business provides a robust six levels of QoS with congestion avoidance based on Assured Forwarding. However, Verizon doesn't allow you to choose which IP precedence values go into each of the 6 queues or set the bandwidth allocated to each queue. Verizon also charges extra for VoIP bandwidth. AT&T and Sprint offer four classes of service. Other carriers offer only three since they are using partner carriers to build a global MPLS service and have to default to the lowest common denominator - High, Medium, and Low.

Furthermore, despite all the carriers offering QoS, except for VoIP (IP precedence 5), none of these QoS models align. The traffic, based on IP precedence number, Verizon puts into its 2nd highest queue is different than what AT&T uses. The next carrier changes something else. Some carriers are starting to set CS1 to a scavenger class to hold P2P and garbage traffic. But other carriers still use CS1 for normal traffic.

This wreaks havoc on customers who use MPLS service from more than one carrier. Many enterprises, especially multinationals and governments, require a dual-carrier strategy. This makes QoS a real headache when the two carriers you choose have completely different QoS models. When traffic flows over the first carrier do you mark traffic one way, but when it flows over the other carrier mark it another? What about when a single traffic session flows over both carriers? How do you set bandwidth levels for each type of traffic when the carriers have different queues? You can see how the design spirals out of control quickly, not to mention the operational support.

What I'd like to see is carriers provide an industry standard 4-level QoS model for MPLS networks. That's not a lot of queues (like Verizon's six) which some carrier backbones cannot support, but goes beyond the overly simply high, medium, and low QoS model. Plus, customer should be able to set the bandwidth allocated to each queue. Which packet markings (IP precedence) go into each queue should be part of the industry standard. Customers can choose to use the standard 4-level or, if they prefer the carrier's proprietary model, customers can choose to use that one. This way customers can simplify their QoS designs when using dual-carriers, but still allow carriers to differentiate themselves with advanced QoS. This makes QoS easier and more customers would use it.

Now, let's see if the carriers start working together on this....

• LANs / WANs