Ethernet’s secret source

Yes, token ring is gone, but it was never a flop. In retrospect, today’s Ethernet resembles more closely the token-ring IEEE 802.5 standard than it does its own original IEEE 802.3 standard.

As Network World turns 20 next month, the editors have asked columnists to opine about some technology flops of the era. A variety of possibilities came up, but token ring was offered up as a prime candidate.

Yes, it’s gone, but it was never a flop. In retrospect, today’s Ethernet resembles more closely the token-ring IEEE 802.5 standard than it does its own original IEEE 802.3 standard.

The official name of the Ethernet standard gets to the heart of it: Carrier Sense Multiple Access with Collision Detection, or CSMA/CD. The name describes how Ethernet works, or how it used to work. The multiple stations would listen to find out whether another station was transmitting. Hearing none, it would transmit. Another station could be doing the same thing, and the two transmissions could collide. When that happened, the stations could detect the collision, wait, and try again.

This approach works when you have a small number of stations, because collisions rarely occur in such cases and throughput is good. CSMA/CD is non-deterministic, which means the performance for any single station is essentially unpredictable. As more stations try to access the same Ethernet segment, there are more collisions, more retries, more collisions.

As CSMA/CD networks get busier, throughput degrades dramatically.

CSMA/CD Ethernet did not scale. To improve performance, existing LANs had to be broken up into smaller ones. Not only did this require leaving unused ports in existing hubs, buying new hubs and potentially using then-expensive router ports to allow segments to communicate, but frequently reworking the wiring closet also triggered considerable downtime. The only way to deliver predictable, good performance was to have a single station on each LAN segment.

Token ring, by contrast, was designed to be a deterministic network. Performance remained good regardless of the number of stations or the activity on the network. Studies showed that token ring could be run to more than 95% of theoretical max – something CSMA/CD could not approach.

CSMA/CD was headed to the graveyard until engineers invented the first Ethernet switches in the mid-1990s. At once, this device provided Ethernet with one station per segment, which gave Ethernet users the deterministic behavior previously available only to token-ring users. Switching removed the segment-scaling issues and nullified the basic CSMA/CD architecture of Ethernet.

Still, Ethernet networks could not indicate priority, although token ring’s architects had built in ways to mark frames. Ethernet architects adopted this feature when it painfully squeezed new fields into the Ethernet frame header to carry virtual LANs and the 802.1p’s eight levels of prioritization.

Ethernet was still without a large-frame capability to stream large data transfers across the network. While today its official maximum size is 1,518 bytes, token ring circa 1988 allowed 16K-byte frames. The IEEE is catching up and taking on extended frames officially. In crucial ways, today’s Ethernet is token ring “under the hood.”

So, when the discussion turns to LAN technology flops, you should think of the real ones: token bus, Arcnet and 100VG-AnyLAN.

What do you think? Discuss Ethernet and LAN technology flops in our forum.