Wonder twins

Communications News, April, 1998 by Tim Hale

Superheroes Gigabit Ethernet and ATM combine powers to battle the evil forces of FUD.

If Superman fought Batman, who would win? It's a silly question, but one we've all heard. And why? Because competition is exciting. Never mind the fact that they're both good guys and that, if they were real, they would probably work together to fight injustice. It's more fun to pit them against one another.

The same seems to be true about Gigabit Ethernet and ATM. But the question isn't "Who will win? It's "How can they work together to fight network congestion?"

There is no doubt that Ethernet rules the desktop. With PCs getting faster and applications demanding more bandwidth, 10/100 network interface cards (NICs) have now become the best-selling type. It's natural to conclude that eventually many of those NICs will be configured to run in Fast Ethernet mode. In preparation for that time, network managers are considering Gigabit Ethernet, with its familiar underlying technology and high bandwidth, for the link between their data-oriented workgroups.

Many understand the savings they can realize by integrating their existing voice and data networks. They know that ATM, with its inherent support for quality of service (QoS), allows different traffic types to run over protected circuits on a single network infrastructure. ATM provides the resiliency and scalability (in terms of both distance and bandwidth) that are essential for connecting buildings across the campus and campuses across the enterprise backbone.

ENTER THE EVIL FUD SLINGERS

The integration of these heavyweight technologies has real advantages and at least one major challenge, but technology evangelist continue to sling Fear, Uncertainty, and Doubt (FUD). Neither technology is a networking panacea in itself, and there are many misconceptions about each that need to be addressed before a winning side can be proclaimed:

* Gigabit Ethernet will travel up to 260 meters over 62.5 micron multimode fiber when the 802.3z standard is finally approved for 1000Base-SX (62.5 micron fiber is common to the vast majority of installations in North America). ATM can travel 2,000 meters (2 kilometers) over the same fiber today.

* Contrary to some reports, Gigabit Ethernet does not support QoS. There is nothing in the standard that specifies this capability. Though add-on prioritization schemes are not excluded by the spec, they are not built in.

* 802.1p and RSVP are not standards yet. They will provide class of service (CoS) traffic prioritization when they are complete, but unless applications and operating systems are written to request specific priorities, network managers will need to manually classify traffic at every network element throughout the LAN. In an integrated network, this task will soon be simplified with management tools.

* Application-level QoS is still a long way off. Even though ATM supports Qos now, there are still only a few niche applications can signal for connections that meet their specific traffic management parameters.

* Although some Layer 3 switches support Gigabit Ethernet interfaces, Gigabit Ethernet switches do not yet support high-speed hardware-based routing. Likewise, Layer 3 switching ASICs provide a vast performance improvement over old processor-intensive routers, but they do not yet have the horsepower to handle the port density of a Gigabit Ethernet switch.

* ATM is a multi-layer technology. LAN Emulation (LANE) provides Layer 2 bridging of LAN protocols, and multiprotocol over ATM (MPOA) will enable Layer 3 cut-through routing. MPOA does not compete with Layer 3 switching.

In fact, it will be possible to implement the MPOA Server (MPS) functionality on a Layer switch. This will provide the high-speed routing of the Layer 3 switch with the added benefit of cut-through connections for long duration flows.

But to realize the performance enhancement of MPOA, switches at the edge of the LAN will also require Layer 3 ASICs. MPOA Clients (MPCs) will spot flows, request short cuts, and reframe packets. This extra work will overwhelm edge switches that are not hardware-enhanced to support MPOA. Using Layer 3 switches to route between ELANs is preferable to early, software-only implementations of MPOA. And implementations that comply with the ATM Forum specification for MPOA (which requires LANE 2.0 and UNI 4.0) are still several months away.

* The so-called "cell tax" is hogwash. ASIC-based segmentation and reassembly (SAR) technology and low-latency, wire-speed switching have rendered the overhead imposed by the ATM cell structure a non-issue.

And the "packet tax" imposed by IP, UDP, and RTP headers applies only when Ethernet is used to carry the small packets required by time-sensitive voice and video traffic. This, too, is a non-issue in an integrated network because isochronous traffic can be handled directly by the ATM core.

ATTRACTIVE QUALITIES

Gigabit Ethernet is attractive because it's simple, fast, and relatively inexpensive to implement. ATM, though scalable, is not typically needed at the desktop. Its place is in the backbone where end-to-end QoS allows multi-service integration and simplifies network management.