The following text is copyright 1997 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.

Do ATM switches dream of gigabit Ethernets?

This gigabit Ethernet stuff is beginning to look quite real. I saw a copy of the 802.3z document proposing a mode for "1000 Mb operation of 802.3" back in January. It was then a full specification. Last month the working group met and seems to have finalized all of the technical details and expect to have a document ready for the first of a long series of approval steps this summer. This document should be very close to the final version, close enough for vendors to start shipping products. The 802.3z group seems determined to avoid the kind of fight that delayed FDDI for many months when trying to determine what the shape of the connector should be.

Now products are beginning to appear with 3Com and Packet Engines among the vendors that have already announced specific devices. I've even gotten word that two or more Ethernet switches with gigabit uplinks will be submitted for the annual pre-Interop Harvard Network Device Test Lab performance test series. I expect that quite a few vendors will be displaying devices (or at least pictures or shells of devices) at Networld+ Interop in Las Vegas in May and will be ready to take your orders soon after. (Though I would get a clause in the purchase agreement guaranteeing free, or very low cost, upgrades to full standards-compliant versions before I'd buy too much equipment.)

The technology itself looks good and simple aside from some semi-peculiar things that have been done to preserve a reasonable collision diameter (network span) for the half-duplex case. At least one vendor has already announced a "buffered-repeater" that supports full duplex links in and out of a hub by using FIFO buffers on the input ports. This permits a quite simple repeating hub design but confines the collision zone to the hub backplane. As long as the buffers are big enough and both ends support 802.3x flow control this should work just fine. Since connections to a buffered repeater are full duplex, the special features that are needed to support half-duplex are not required.

All of this activity in the gigabit Ethernet camp must be causing some restless nights in ATMville. In most places ATM has already lost the battle for the connections to the desktop to switched 10 Mb and 100 Mb Ethernet technologies. Now the imminent arrival of gigabit Ethernet devices is threatening the clear lead ATM had in mindset for corporate campus backbones. Even ATM's strong quality of service story becomes less important at these speeds since it does not take very long to clear any queued up data. From this vantage point it looks like ATM as an important technology is retreating to the pure WAN environment, where bandwidth is not so cheap, and to those sites where a large WAN net is tightly intertwined with a campus backbone. And note the faint drums in the distance, initial discussions have already been held for extending Ethernet to 10 Gb.

disclaimer: Note that people at Harvard often hear drums in the distance, to one level of reliability or another, in this case the perceptions are mine.