We Should've Listened
Back in the mid-eighties, I worked as an in-house consultant for a global entertainment firm, helping the conglomerate move one of their regional offices from mainframe-based applications and terminals to PC-based systems, linked together on the company's first full-time production LAN. As could be expected, this proved to be the source of many battles, most of which revolved around the products and technologies that would become the "standards" for the firm.
Surprisingly, the most arduous of these battles was over infrastructure topology. At the time, commercial Ethernet was still fairly new, although most of the existing vendors were supporting it, or at least had plans to do so. On the other hand, IBM was pushing Token Ring hard (or rather, IBM was ignoring Ethernet hard), and this company had lots of IBM mainframes and minicomputers scattered across the globe, giving that technology an upper hand in the decision process.
During the months of debate that ensued, I was always surprised how IBM's people didn't focus too much on their obvious strengths in integration, but instead would talk about Token Ring's features such as large frame sizes and managed-access. IBM's people would always point out how important these features would be once multimedia-on-the-LAN became commonplace. One day, they said, we'd be glad we spent the extra money on the better solution.
None of this mattered of course, since we were really only looking at the SNA factor. Maybe they knew this and were just trying to help us justify the purchase. Who knows. One thing was certain, though: None of us ever expected network-multimedia to take off in a big way, particularly when you consider we were using 286s or first-generation 386s coupled with low-end monitors on the desktop. We could barely get Windows 286 to work, never mind network-multimedia.
The odd part about this whole experience was that most of the Ethernet vendors agreed with us. For the most part, the vendors at that time were primarily pushing low-cost and wide-availability as Ethernet's primary features. And with the exception of Bob Metcalfe, nobody seemed willing to argue Ethernet's technical advantages for very long. In the end, most of them agreed that Token Ring was probably the better technical solution, but they also argued that it was more solution than our problems required, and that we'd be better off just buying into a commodity technology like Ethernet.
We agreed with this approach, at least in our hearts. The commodity-value concept was our primary reason for choosing Windows over the Macintosh for our desktop clients, and was our motivation for eventually switching from LAN Manager to NetWare. Using market-leading technology has indisputable advantages in terms of support and application availability.
But in the end, none of this mattered. We needed SNA to the desktop, and Token Ring was the most-viable solution to that problem at the time. The technical supremacy of Token Ring's design - particularly in the face of network-multimedia - was absolutely irrelevant to our decision to go with it.
Oh, How Things Have Changed
Now, fast-forward to the present, where network-multimedia is becoming almost commonplace, with technologies like Voice-over-IP and LAN-based video-conferencing proving to be viable alternatives to their circuit-based counterparts. And I'm not sure you can even buy a major-brand computer these days without getting speakers and a microphone, or at least a sound card anyway. All of a sudden, network-multimedia is not only possible, but it's proving to be feasible as well.
Yet, network-multimedia has some pretty stringent requirements in order for it to work successfully. In particular, these kinds of technologies require large amounts of dedicated bandwidth at the end-point stations, and integrated prioritization services within the network itself. The original Ethernet - with its shared-access coaxial network that depends on luck in order to function - is woefully unsuited for the job. Token Ring, on the other hand, is infinitely better-suited to the task. IBM was right after all. Who'd have ever thought it.
Of course, Ethernet has changed over the years as well, and the current offerings look more and more like Token Ring every day. Indeed, I would even go so far as to say that today's Ethernet is Token Ring, for all practical purposes. Here are just some of the similarities:
- Managed-Access: With the advent of Ethernet switching, Ethernet's biggest problem - the shared-access medium - was eliminated. Rather than hoping that nobody else needs to transmit data, Ethernet switches allow everybody to transmit as needed, with the switch providing congestion- and traffic-management as needed.
- Prioritized Traffic< Another major advancement in Ethernet technology is 802.1Q, which now allows Ethernet frames to carry prioritization markings directly in their headers, thereby allowing the Ethernet switches to prioritize traffic on a per-frame basis. Thus, urgent data can be clearly marked and identified, allowing database updates to go through even when the network is saturated with lower-priority VoIP traffic. Guess where this idea came from.
- Large Frames: Companies like Alteon have recently started lobbying for a new "Jumbo" Ethernet frame, capable of carrying nine kilobytes of data instead of the 1.5 kilobytes normally found on Ethernet. Larger frames are great on fast, high-speed networks, since they allow more data to be moved in less time, and use substantially less overhead on the end-point systems. Of course, this has been known in Token Ring circles for years, where frame sizes as large as 64 kilobytes have long been available.
- High Prices: Although low-end, no-name Ethernet NICs are still cheaper than Token Ring cards, the cost of a good Ethernet NIC - especially when combined with the per-port cost of a 802.1Q-compliant, Jumbo-frame switch - is more expensive than any equivalent Token Ring solution. For the same basic functionality, Token Ring is actually cheaper.
Price, There's The Rub
This last point is the most interesting, I think. While many folks have long said that the low-cost nature of Ethernet's meek design was a better bet over the expensive nature of Token Ring's too-much-solution approach, what we're seeing today is that a solution which is capable of efficiently handling network-multimedia makes Token Ring the better bargain. Building out an Ethernet network of equal strength actually costs more, and with significantly less functionality in many key areas.
This is especially true when you look at costs spread out over an extended period. For example, the entertainment concern I described above shelled out a substantial amount of coin to build a significantly-complex Token Ring network that went to every desktop in the organization. However, they only had to build this network once. Conversely, the shops who went with the "cheaper" Ethernet have since rewired their networks two or three times in that same period, and they still don't have switched-access, prioritized traffic and large frames to every desktop. They'll have to rewire everything at least one more time (and probably twice) before they've even caught up with where we were a decade ago.
Now, I don't want to come across like I'm all for Token Ring. I use Ethernet here and wouldn't dream of trying to rely exclusively on Token Ring, but that's because my network is relatively new and small, and I'm not in need of certain elements that Token Ring offers. Indeed, there's lots of places where these advanced characteristics just aren't needed. Nobody needs prioritization and Jumbo frames on their inkjet print server.
And to tell the truth, Ethernet has had lots of advantages over Token Ring. The most-compelling advantage of course is the 100 MB/s access speed, and more-recently, the full-duplex transmission capabilities that switching has allowed for. However, both of these are proving to be short-lived advantages, with a variety of Token Ring vendors offering these same technologies (just as Gigabit Ethernet starts to take off).
But what I am saying - and this is important - is that what we all thought 10 years ago has proven to be wrong. We do need better networking infrastructures than Ethernet. The demands that technologies like network-multimedia place on your infrastructure - high-speed dedicated connections and prioritization in particular - aren't any cheaper or easier to solve with Ethernet than they are with Token Ring. In this regard, there is very little advantage in using one or the other when you're trying to bring a robust infrastructure all the way down to the desktop.
Most organizations are going to have to overhaul their networks (those that didn't choose Token Ring, anyway), due to the demands from VoIP and other technologies. Maybe the best approach from here on out is to examine the true costs of building a network that can handle the load, and then choosing the technically-superior solution that can get you there in the long run. Rather than thinking about this quarter's budget, think about the cost benefits over the next few years, and get it right the first time. Maybe, just maybe, some of those vendors are right.