One Ring to Rule them all
10Mbps, 100Mbps, 1Gbps, 10 Gbps, _______
So what is the next number in this series? If my 12 year old son was asked to complete this series in his math class, I’m pretty sure that he would be able to put in 100Gbps, without even knowing what bps means.
For those of us in the IT industry who know that it mean bits per second, we also know that the next number in this series is 40Gbps. So why has Ethernet fallen off of it’s power of 10 progression?
With Fiber Channel, successive generations have seen a doubling of bandwidth, 2G, 4G, 8G and 16G; likewise for Infiniband, single data rate, SDR, then DDR and then QDR. The mighy Ethernet has managed to increase 10 fold with each generation.
To do this, Ethernet has “cheated” a bit. The transition from 100Mbps to 1Gbps, Ethernet changed from a multi-point connection, to a point to point connection, that helps with the signal integrity a great deal. Many felt at the time 1Gbps would only be practical with optical cables, but vendors have found a way to make copper links practical. Even 10Gbps Ethernet can be transported over copper for a limited distance. In this case the enabler was very sophisticated DSP in silicon which performed channel equalization and careful waveshaping. Certainly the practical distance over copper medium is getting shorter, but we have gotten further with copper than most would have predicted in the days of Fast Ethernet.
Although 100Gbps is certainly on the roadmap for Ethernet, why the pit-stop at 40Gbps? Is this copper’s last stand?
source: alysion.org
Each successive generation of Ethernet starts out expensive and begins market adoption where the new speed provides the highest value. As the technology is adopted, the volume goes up, price goes down, enabling wider spread adoption. It has been the volume of Ethernet which has made it the king of interconnects. In the data communications world, if there is to be “one ring to rule them all” it would be Ethernet. Ethernet is even encroaching on Sonet territory in the telecommunications world. No other technology has been able to win against Ethernet. Infiniband and Fiberchannel have carved out their own market segments, but I’m wondering if the data communications market will sustain three wired interconnect technologies. Ethernet has a habit of “learning” important tricks from these competing technologies to encroach on their markets, the addition of RDMA to Ethernet is a threat to Infiniband, and FCoE is a threat to Fiberchannel. Infiniband won the interconnect wars in the High Performance Computing market against competitive technologies like Myrinet and Quadrics, in this domain, high bandwidth and low latency rule, but 10Gbps Ethernet is closing the gap on bandwidth and latency with Infiniband. Although Infiniband will likely always be superior in performance as it is designed specifically for the high performance market, market disruptors innovate from below with less capability at a much stronger value proposition. Besides the port cost advantage that Ethernet brings to the table, it also allows for a simplification of networks in the data center. Why have 3 interconnect technologies, each with their own switching gear, protocol standards and IT management disciplines. Even if you want to keep your storage and cluster networks physically separate, there are still economies to be gained by using the same technology as the base.
With all this, we haven’t even opened the subject of Ethernet in the wireless domain. So will Ethernet be the technology to unite all of the IT world?
I’d like your thoughts on why the Ethernet pit-stop at 40G, as well as your thoughts on how many wired interconnect technologies the IT market will sustain in the long term.
May 5, 2010 - 8:11 am
While I cannot really comment on 40G (I don’t know enough about this area of IT), I still find it hard to believe that wireless N connections for the home network at 300Mbps is faster than the wired one at 100Mbps (theoretically, and dependent on distance from router).
May 5, 2010 - 9:24 am
That’s a good point Chris. These are ‘wirespeed’ bit rates, the number of bits per second at the physical interface, it doesn’t matter whether its copper, optical or radio. The actual payload ‘throughput’ is quite a bit less; it depends on the number and types of protocol layers between the physical layer and the application layer which each add their own overheads eating into the performance; and on the quality of the channel; noise in the channel for instance could trigger a lot of retransmissions if you’re using TCP which will really hammer your throughput. Noise increases with distance for all media. Shared medium like radio will also spend some of their bandwidth on negotiating time slots for participants; you probably have more that a few devices in your wireless N network.