When I go to Las Vegas, I enjoy the fine food (no, not the buffets), I see some shows, and I watch the people, but I don't gamble. I'm just not a gambling man. That's a shame, because I know of one technology that's a sure bet to win all contests: Ethernet.
I have watched Ethernet triumph over competing technology for years now. Ethernet not only beat but pretty much eliminated ARCNET and Token-Ring from the face of the planet (I know, I know, some Token-Ring is still out there). Ethernet hasn't outright destroyed InfiniBand, but it has definitely pushed it into a corner. And now Ethernet has its sights set on Fibre Channel.
What's amazing about these victories is that Ethernet isn't necessarily the technically superior solution. For example, Token-Ring was faster, and its deterministic protocol let bandwidth be more easily allocated and predictable. Similarly, InfiniBand has a better balance of low latency and high bandwidth. Nonetheless, Ethernet marches on.
I've heard many theories about why Ethernet has been so successful. Some argue that Ethernet is always the lower-cost solution. Others argue that we have become so comfortable with Ethernet that it's impossible to displace. I'm unsure. I think it's some mixture of these and other factors, but I am sure that I wouldn't bet against Ethernet . . . if I were a gambling man.
Traditionally, Ethernet has been a LAN technology targeted at interconnecting servers, communications equipment, and of course, client devices (e.g., PCs, laptops, etc.). Ethernet generally was not used for storage connectivity or for server-to-server connections in high performance computing (HPC) clusters. Three barriers kept Ethernet from addressing these usage cases: bandwidth, latency, and packet dropping.
New Ethernet standards have addressed these concerns. For the bandwidth problem, we have 10 Gbps Ethernet, which puts Ethernet on par with (or makes it better than) any other interconnect. TCP/IP Offload Engines (TOE) help improve latency in Ethernet, making it more competitive with InfiniBand and Fibre Channel. Last, a new standard for lossless Ethernet eliminates the packet dropping that occurs in normal Ethernet networks (those packets get recovered eventually). This last item is crucial for storage traffic.
Two additional developments also make Ethernet more attractive to the storage world: Internet SCSI (iSCSI) and Fibre Channel over Ethernet (FCoE). iSCSI lets servers and storage devices communicate over end-to-end Ethernet connections. FCoE is a little more complicated; using FCoE, storage traffic leaves a server over an Ethernet link and then travels to a gateway device that reroutes the traffic over a Fibre Channel link to a storage device.
All these advancements (and more) are packaged under the moniker "Data Center Ethernet," and although I hate the name, I must say the combination of technologies is "hot."
As much as I like the direction it is heading, I do dislike a few things about Data Center Ethernet. First, if you want all the benefits of Data Center Ethernet, you need to replace your cabling, your network switches, and the network adapters in your servers and storage devices (or replace servers and storage devices in some cases). No problem, right?
That brings me to my second point: the high costs associated with Data Center Ethernet. Yes, the costs are coming down, but that means they're still up. Speaking of "up" (as in "up in the air"), not all the standards have been formally approved, and details such as cable and connector types are still disputed.
Again, I like the direction of Data Center Ethernet, and I wouldn't bet against it (if I were a gambling man), but having to replace core infrastructure with expensive and not-yet-standardized technology is "not hot."
Sean Chandler is a computer and network consultant who has nearly 30 years of field experience. Astro, a border collie with more than 40 dog years of data processing experience, provides technical support to his master, Sean.
I've heard the phrase "what happens in Vegas stays in Vegas," but I rarely hear about how technology has similar boundaries. Simply put, some technologies available in Europe and Asia are unavailable in the U.S. One of my master's European friends shared one of these technologies with him — a cell phone jammer. This little device is small enough to hide in a pocket but powerful enough to jam cell phone signals within a 15-meter radius. Think about it: Tired of people near you talking in restaurants or theaters? Turn on the jammer. Tired of chattering teenagers standing in checkout lines behind you? Turn on the jammer. Too bad this device is unavailable in the U.S. Even worse, too bad it's actually illegal to have one in the U.S. — which my master discovered when he tried to import one!
— Astro
Links:
[1] http://systeminetwork.com/author/sean-chandler