New ATM service targets IP networks

ATM is getting a new class of service tailored specifically for IP backbone nets. The guaranteed frame rate (GFR) class of service will support frame-based traffic - both IP and Ethernet - better than ATM's current best-effort services do.

GFR is being jointly developed by the ATM Forum and the Telecommunications Union- Telecommunications. If all goes as planned, large enterprises and carriers will see GFR products next year.

GFR will be ATM's sixth class of service and the first new traffic specification to emerge since the ATM Forum's Anchorage Accord of 1996. The Anchorage Accord froze the development of new standards for two years.

At the heart of the Accord was the decision to group standards into foundation-layer specifications and extensions that will be built on top of those specifications. Foundation-level components included the Physical Layer specification, Traffic Management 4.0, Signaling 4.0 and Private Network-to-Network Interface 1.0. Extensions included specifications such as Multi-Protocol over ATM and LAN Emulation 1.0.

GFR falls under the auspices of Traffic Management. GFR will optimize the handling of packet-based LAN traffic that otherwise relies on unspecified bit rate (UBR) service across ATM backbones.

Today's UBR is a best-effort service, similar to frame relay's zero committed information rate (0-CIR). UBR carries IP or other traffic coming off an Ethernet LAN when network bandwidth happens to be available. UBR was developed three years ago as a way to accommodate legacy LANs in the early ATM market.

Available bit rate (ABR) is another ATM service meant to provide a greater measure of guaranteed packet performance over ATM backbones. But ABR remains difficult to implement between routers in an ATM network and is not considered a practical service option.

Better packet service is imperative because today's ATM requirements are focused more on IP traffic than they were when UBR and ABR arrived. ATM is being used in large enterprise, carrier and ISP networks to consolidate and extend IP services over the wide area. ATM brings scalability and greater flexibility to the IP core, and its importance will increase as new IP services (including packet voice and video) roll out.

GFR is being developed to provide the exact quality of service required by all types of IP traffic. The intent is that GFR should become the enabling ATM class of service (CoS) for IP's backbone growth.

Applying to both Layer 2 and Layer 3 protocols (Ethernet and IP), GFR will provide throughput guarantees at the frame/packet level and will deal more fairly with such traffic at the ATM level when congestion occurs.

The technical challenge is to find a better way of not dropping packet-based traffic. Designers plan to support three levels of GFR conformance, with fairness rather than priority being the key. GFR will help to achieve better traffic shaping and fairness among competing IP traffic streams in a crowded ATM backbone. Ideally, GFR could be implemented as a software upgrade to existing ATM edge devices. But it may be that GFR entails new conformance algorithms requiring hardware upgrades.

These input algorithms, which express ATM's classes of service, are implemented where traffic enters a switch. They keep tabs on the incoming traffic and will only admit cells that conform to the traffic profiles established for each circuit.

But ATM performance hinges more on the output side of a switch, where throughput is measured. For example, ATM traffic management and congestion control happen on switch output so that if there's congestion in the network ahead, traffic is buffered before it leaves the switch.

In other words, GFR alone is not enough to ease IP traffic through an ATM backbone. GFR will define a better CoS for IP handling, but to succeed GFR will require the best traffic management that ATM switch vendors have to offer. It does no good to let IP traffic into a switch more efficiently if the traffic cannot be managed fairly on the way out.

The more recently developed adaptive-discard approach to traffic management is a better technology for ATM's IP backbone role. Adaptive-discard is particularly well-suited to GFR's requirements.

Adaptive-discard refers to dynamic, two-dimensional traffic algorithms that can adapt to the changing fairness of network conditions. The technology reacts not just to bandwidth availability and burstiness, but to the fair allocation of throughput and buffering among competing traffic streams.

In comparison, per-VC management is defined by static thresholds that do not respond as amenably to bursty traffic. Adaptive-discard gives ATM greater flexibility for GFR to work with, especially at the WAN edge where flow control and fairness affect network performance.

The GFR CoS will grant more distinction to IP streams being carried in an ATM network. The specification is being written to allow more fairness in the way packet-based traffic is handled. Enhanced ATM traffic management based on adaptive-discard will be the factor on which GFR thrives.

Catanzariti is manager of technology planning at Network Equipment Technologies, Inc., a network hardware vendor in Redwood City, Calif. He can be reached at sergio_catanzariti@net.com.

The quality-of-service quandary:
Network World, 4/13/98

Available bit rate services deliver quality of service in ATM networks:
Network World, 9/15/97