Wednesday, August 22, 2012

Ethernet's future: How fast is fast enough?

Your PC may not need a faster network, but overall data usage on the Web is doubling every year. The big question: Can the venerable Ethernet standard handle a terabit per second?

 Slow network speeds got you down? On Monday, computing experts will announce they're tackling the next speed bump for the venerable Ethernet standard. But don't expect to find the new speed option on your next computer's feature list. The standard, to be produced by the Institute of Electrical and Electronics Engineers (IEEE), will likely reach data-transfer speeds between 400 gigabits per second and 1 terabit per second. For comparison, that latter speed would be enough to copy two-and-a-half full-length Blu-ray movies in a second. In contrast, your laptop today probably maxes out at a mere 100 megabits per second -- maybe a full gigabit per second. 

And that's assuming your laptop even has an Ethernet port in the first place -- MacBook Airs don't, and the trend seems likely to spread as thin laptops catch on. Many people probably don't even know how fast their Ethernet connections are, because the real bottleneck these days for most users is their broadband connection to the Internet, not the Ethernet connection that can links personal computer to the local network at home or work. But a faster Ethernet standard still matters to ordinary people. The companies at the other end of the Internet connection -- the Facebooks, the Googles, the telecommunications and financial services firms -- are experiencing unrelenting growth in network capacity needs. And if they can't expand economically or add new features, the Internet either slows down, gets less useful or entertaining, or gets more expensive to use. "The bandwidth associated with core networking was observed, on average, to be doubling every eighteen months," engineers concluded in a July IEEE Ethernet bandwidth assessment (PDF). Thus, the engineers have decided to convene a new group to figure out how to satisfy that need. "For 2015, we expect the bandwidth that needs to be supported to be 10 times what it was in 2010, and in 2020, 100 times what it was in 2010," said John D'Ambrosia, chair of new Higher-Speed Ethernet Consensus group that will lay the groundwork for the actual standard. A big part of the group's work will be figuring out whether 400Gbps or 1Tbps is a better approach, he said. Right now, companies that build the networking hardware tend to favor the 400Gbps option, but the customers who use the network hardware favor the 1Tbps option, which is 2.5 times faster. The key question will be figuring out whether terabit Ethernet is economically practical for the hardware companies involved, he D'Ambrosia said. 

 "People realize 400-gig Ethernet is technically and economically feasible. When you look at terabit ethernet, it's driven solely by demand," D'Ambrosia said. "People know there's a tsunami of data coming. It's basic math: terabit is more than 400-gig, so we want a terabit. That's nice, but one has to worry about the technical feasibility and the economic feasibility." Whatever approach is used, it will likely work by bundling multiple physical connections together into one higher-speed virtual connection. That means individual connections don't have to be as fast. But there are cost limits to bundling multiple connections together, since each link needs its own components, and that can get expensive. For example, aggregating 25-gigabit connections into a group of 16 would get to 400Gbps, and 40 would get to 1Tbps. Doubling that so data could travel in both directions at that speed would mean 80 connections, and that would be "not pretty," D'Ambrosia said with some understatement. Especially with copper connections. Cables that bundle numerous strands are thick, unwieldy, and heavy. In supercomputing centers, there are facility issues with data centers when cables get too thick. Some contributors to the bandwidth assessment report said they've had to "go in and reinforce our high-performance computing [data centers] because of the sheer weight of copper cables," D'Ambrosia said. "A cable 2 inches in diameter is incredible." 

The Ethernet port on the side of PCs uses an RJ-45 connector and copper cables. But where customers use 100Gbps Ethernet, both fiber-optic links and copper technology appears. Copper tends to be cheaper, but fiber optics can reach farther. With today's 40Gbps and 100Gbps standard, copper cables can be a maximum of 5 meters long, and making them shorter raises practical concerns. The objective of the new group will be to build a consensus about how fast the next-gen Ethernet should be. Terabit Ethernet sounds great, but "it's not just a matter of what you want. 

It's how much you're willing to pay for it," D'Ambrosia said. Hashing out the priorities and limits means that the actual process later of designing the next-generation interface will go faster, he predicted. With the bandwidth assessment in hand, though, laying out the need for a standard 100 times as fast as today's by 2020, the first stage of the debate is over. "I suspect this is going to be a very fast-moving project," D'Ambrosia said.

By: Stephen Shankland

No comments:

Post a Comment