11-10-2010 07:23 AM
I loved growing up in the '70s. It was fantastic. It was goofy. And it was also the time when we began seriously experimenting with...
You guessed it - networks!
We were all so excited to get our first Ethernet networks. There would be Ether tapping parties where, if successful, we would celebrate one more host being able to communicate across the latest and greatest. The proof of the successful vampire tap was the pile of cladding and plastic that was proudly displayed on the table along with the tapping tools.
And so we started to get a little crazy with topologies.
We discovered that because Ethernet frames don't have time-to-live or hop count fields, poorly designed networks with loops would show 100% utilization with no hosts transmitting. The fix? Just unplug something for a bit, let the frames run out on the floor, and the network would quiet down. Some of you may think I'm joking about that, but the old timers like me will remember that I'm not! (Ring-Ring! "Hello?" "Hi Chip, the network is running slow." "Hang on..." Unplug something, count to 5 plug it back in. "How's that?" "Much better, thanks!") But having to actually get up and unplug something was a hassle, and voila - the spanning tree protocol was born.
What a great idea, eh? Instead of having to physically unplug stuff to make it work, we could just logically unplug it. Sure, you don't get to use all the connectivity in the network, but if something goes wrong, the network will automatically fix itself. Eventually.
And besides, just like we got these networks going during this decade, the '80s will surely fix that, right? Right?
Instead of fixing the problem, over the decades, we've made it worse. We could eliminate the need for spanning tree, or we could just make the spanning tree protocol run faster. We could eliminate the need for spanning tree, or we could just run multiple instances of it. We could eliminate the need for spanning tree, or we could invent technologies like multi-chassis trunking to make it look less like a spanning tree while still being a spanning tree.
Or we could just eliminate the need for a spanning tree.
And we did that. Over 10 years ago. We had a chance with Fibre Channel to take our experiences to figure out what to do and what not to do. And one of the things NOT to do was have a need for a spanning tree. And it works very well! And, for some reason, those improvements never made it to Ethernet.
Welcome, Ethernet Fabrics! Welcome, technologies that are over a decade old, but are finally making it over to the Ethernet side.
The tough part is going to be trying to convince old guys like me to drop technologies developed 40 years ago in favor of technologies developed just 10 years ago to get rid of complicated problems we created, then fixed with complicated solutions. Goofy, huh?
11-10-2010 08:15 AM
As a fellow "over 40" I can "get behind where you're at", as the saying goes. Indeed, improvements in key technologies, such as Ethernet, do happen. But innovations (rather than refinements of an existing technology like STP) require problems large enough that small incremental refinements won't cut it any more.
As I discussed in , the risk/reward ratio for classic Ethernet sifted dramatically with server virtualization. There is too much money at risk (amplified by a global recession) to slow virtualization adoption. So, when an existing network technology (classic Ethernet and STP) can't keep up, it's ripe for a major innovation.
As you correctly point out, Ethernet fabrics are not risky in the sense we have a very successful (more than a decade) "proof of concept" if you will called storage area networks (SANs). They rely on fabric technology for their success (like doubling the traffic flow every year as storage doubled, six nines of availability, and plug-in scalability of inter-switch traffic). Because they are robust and reliable, much of the world's data flows through a switched fabric connected to storage. I think the technical risk of fabrics is pretty low.
But, humans, particularly the subspecies "IT administrator", are risk adverse people. They move cautiously, as well they should, when you consider the economic value of the applications, storage and client traffic that moves over their networks. It's my experience that "administrators" exhibit tribal behavior breaking into the server clan, storage clan and network clan so they don't always interact a lot.
So, my proposal for the network folks who understand the risks they face with today's classic Ethernet, go down the hall and talk with the leader of the "storage clan" and ask her to explain how storage network fabrics work. For a lot of companies, network administrators can perform a lot of due diligence about Ethernet fabrics right in their own building.
11-10-2010 12:45 PM
"Welcome, Ethernet Fabrics! Welcome, technologies that are over a decade old, but are finally making it over to the Ethernet side."
I could suggest that it sounds a lot like switched Token Ring Networks.
Good to see ethernet finally catch up !
11-10-2010 01:09 PM
Well, I don't think Chips comments means an Ethernet fabric is like Token Ring. You could think of an Ethernet fabric as taking the "switch fabric" that connects all the ports in a classic Ethernet switch and extending it between switches in one rack, many racks or even across network tiers such as the access and aggregation tiers.
So I think its more correct to say an Ethernet fabric is like a storage fabric in that many of the networking innovations introduced were brought forward and updated to be suitable for Ethernet traffic.
And I agree with you, it's good to see Ethernet finally catching up. One of the strengths of Etherent is its adapability to change.