Cisco Blogs

Why Ethernet Wins…

May 20, 2009 - 11 Comments

OK, so the title is meant to be a bit ambiguous, as much a question as it is a statement–I’d love to hear your thoughts on this. I have been pondering this a bit since I am prepping for my CCIE re-cert and its kinda fun looking back at how much things have changed in 15 years. It also seems everyone and their brother wants to sell Ethernet switches these days. Companies that have never sold Ethernet are now all over it and companies that have repeatedly walked away from the Ethernet market keep coming back. So, what is it about Ethernet that makes it the networking juggernaut? A lot is always made about the economics of Ethernet, but saying Ethernet wins because it is cheaper is a little too simplistic–hey, LocalTalk was cheap, but you see where that went. Instead, I think Ethernet continues its success because it manages to stay the most cost-effective solution. How does this happen? The first reason is that Ethernet has proven to be infinitely extensible–over the last couple of decades we have managed to incrementally evolve Ethernet to successfully support demanding traffic such as voice, video, and now storage. At the same time, you can get speeds from 100Mb Ethernet to 10Gb Ethernet and in a couple of years, 40Gb and 100Gb Ethernet. So, one secret seems to be that Ethernet offers just-in-time innovation–you don’t have have to buy more capacity or capability than you really need without worrying about investment protection. The corollary for this is that you don’t end up paying for innovation that you don’t want or need, yet you get to benefit from economies of scale created by those who do. In 2006, a 10GbE port on a modular chassis had an ASP of $4,667 and about 226,000 ports were sold to folks who needed them. The volume over the intervening years has given vendors the scale they need to bring down the forecasted ASP for this year to $2,606 with a forecasted volume of 762,000 ports (by the way, these numbers are all vendors, worldwide, courtesy of Dell’Oro).I think the second factor is what I call familiarity. While, every once in a while, someone will announce a new protocol or technology that is interesting, a cool technology, by itself, is seldom compelling. Ethernet represents a comprehensive solution–a well understood technology with a broad selection of operations and management tools and an immense body of operational expertise behind it. In short, its a known quantity. New technologies suffer from this catch-22: a lack of this kind of ecosystem is an inhibitor to adoption, but the ecosystem will not get built unless customers are willing to take the risk with a new, unproven technology, which, these days, most folks are not. You would think perhaps there is some opportunity to build a beach head in a tightly focused area, but even that does not seem to be the case. Take the case of the Top 500 Supercomputing sites–you would think this is a stronghold for a protocol like Infiniband, but 56% of the top 500 sites are using Ethernet, roughly twice as many as are running Infiniband, and the use of Ethernet seems to be increasing.Finally, I’ll toss our the use of unshielded twisted pair (yeah, I know LocalTalk ran over twisted pair too). Having been personally traumatized by the nightmare known as Thinwire Ethernet. 10-BaseT and its UTP based cousins provide a physical layer that is cheap, easy to install and reliable. I think the days of copper are numbered in the data center, but I think it will continue to rule in the campus, and, in any case, UTP is representative of Ethernet’s reputation as a low-maintenance protocol. By default, Ethernet tends to just work–you take a couple of switches, some random hosts, and some RJ45 patch cables and hand them over to my 11 year-old and you will have a functioning network in about as much time as it takes to cable everything together and power it on.While there are a whole host of factors behind Ethernet’s success, these are top of mind for me right now–agree, disagree, what did I miss?

In an effort to keep conversations fresh, Cisco Blogs closes comments after 60 days. Please visit the Cisco Blogs hub page for the latest content.


  1. Can I help the winner drink? ;)

  2. Greg:How come we always end up talking about FCoE and SCSI? :) I'll tell you what, I will put away a nice bottle of Bourbon (or single malt if you wish) and let's revisit this in five years--if iSCSI is winning in the Enterprise data center, I'll ship you the bottle with a nice note. :)Omar

  3. Greg:There is a bit of a ring topology to your logic here :)--you cannot simply say Ethernet won because there was so much of it. Ethernet got to that point because it did what people needed without doing things they did want while keeping things affordable. Based on your logic, we should all be SNA/Token Ring shops because IBM used to have 70% of the enterprise networking market.I agree, FDDI was a great technology, but lets face it--in the early 90s, how many folks needed VLAN tagging, jumbo frames, or the like. It was hard to cost justify FDDI in most scenarios between the cost of the equipment and the fiber if you went that route. By the time we needed to go faster, FE or GbE was an easier/cheaper path--again, it allows folks to buy just what they needed. Yes we are adding back in things that were originally found in TR, FDDI and even ATM (as AKG points out), but now its because we need them are customers see value in the investment.Features = complexity = cost. There is a sweet spot--you need enough features to be useful, and enough features to differentiate, but not so many features that you price yourself out of the market.Omar

  4. @akg--a fair point. One of the reasons that I think Ethernet is successful is the Borg-like willingness to borrow from other technologies, thus, limiting the appeal of new technology.

  5. OmarYou are correct, FDDI was ahead of the curve. Which proves my point, Ethernet got volume because it was easy and cheap, not because it was better designed or more able to scale. Ethernet remains dumb and stupid as it ever was and still doesn't offer good technologies. And we are stuck with it. We have designed around Ethernet shortcomings for years and extended where we could. Don't pretend that this is a good thing, its just living with the problem. Oh yes, in networking, cheap ALWAYS wins. That's why FCoE will takeover from FC, and iSCSI will take over from FCoE. Come back in five years and check it out. greg

  6. Cheapers does not always win. WALMART music store is cheaper then Apple iTunes. Samsung MP3 are cheaper then Apple iPods. Cheaper does not always win. Cheaper and SOMETHING (ease of use, ease to deploy, good design) wins. You need to have Cheaper and one other thing and that wins. Ethernet is CHEAPER and EASY to DEPLOY.

  7. I completely disagree. Ethernet won because it was the cheapest and lowest technology. Let's take a moment to remember just one technology: FDDI had redundant paths, dual homing, QoS, jumbo frames, VLAN tagging, high speed and more from the very beginning. And this was a long time before Ethernet.Ethernet won by being cheap, easy to understand and featureless. This meant large volume which drove the price down even further. As the years go by, we are adding back all the features that we once had. As an industry, we have been LUCKY that Ethernet has been able to be extended, it certainly wasn't by design. History may always be written by the winner. But eulogising Ethernet doesn't wash. Its good enough to work, we've been lucky to extend it. It's only success is that there was so much of it, that it became the de-facto choice. Its called Ethernet Economics"" and can be summarised as ""the cheapest technology always wins""."

  8. @Omar. Agreed. I've been called an ATM bigot :) because I thought ATM in an enterprise/campus environment worked well, even down to the desktop. However, the industry perceived it as too complicated and/or too expensive. Now, we've got many of those f

  9. Ethernet wins because it HAS won. Etherent is the ultimate meme.

  10. Not only does Ethernet offer just-in-time innovation"" but it provides 'just-good-enough' technology. I don't want to start a religious war of Ethernet vs (insert favorite technology here), but Ethernet has, is, and will steam-roller over technologies that were, in many respects, much better. Frame, ATM, SONET, Infiniband....all had and continue to have their merits. Ethernet may or may not be better, but it is good enough to do most of what we through at it."

  11. Indeed.. Metcalfe's Ethernet 1.0 has evolved to Ethernet 4.0 (with CEE/DCE, 40G/100G,...)