Moving to the motherboard
Intel's Waxman also said that he's starting to see 10GbE controllers placed directly on motherboards. As this trend continues over the next few years, he expects 10GbE to be standard equipment on many servers.
Bill Dicke, HP's interconnect strategy manager for blade systems, reminded us that that company's ProLiant BL495c G5 Server Blade, released last September, already includes a dual-port 10GbE LAN-on-motherboard (LOM) controller. According to Dicke, the BL495c is the "First blade server in the industry, to my knowledge, to use 10GbE Flex-10 LOMs."
Considering that Virtual Connect Flex-10 is an HP technology, he's probably right. Flex-10 allows you to divide a single physical 10GbE connection into four virtual connections, each tunable to different bandwidths in 100Mb increments. You can find a video of HP's take on Flex-10 here.
Dan Tuchler, BLADE's VP of product management and strategy, say that another factor behind LON adoption is the stabilization of the PCIe Gen2 standard, which offers increased throughput.
BLADE's President and CEO Vikram Mehta says that the 10GbE ecosystem is ready for mainstream adoption with 10GbE on the server motherboard and affordable SFP+ direct-attach copper cables for the typically short-distance interconnects between server and top-of-rack switches.
Mehta's company's RackSwitch family of top-of-rack switches are aimed squarely at server-virtualization installations.
Mark Hilton, HP ProCurve's director of technical product marketing, says that the the industry's move to server virtualization is fueling the adoption of 10GbE. According to Hilton, one virtual server can get by quite nicely with a one-gigabit connection, so 10GbE can represent 10 virtual machines through a single pipe.
Hilton agrees that 10GbE adoption is increasing, that prices are dropping, and that most of the major technical challenges have been overcome. "I don't see a lot of technical problems at this point," he said. He pointed to HP's ProCurve 6600 Switch Series, which was introduced this January, as an example of his company's 10GbE top-of-rack offerings.
Hinton also believes that the power constraints that previously held 10GbE back are being overcome. Although fiber is a lower-power interconnect than copper, advances in less-expensive copper 10GBase-T technology continue. Hinton sees more power reductions in the future for 10GBase-T, saying "Once we get the power consumption below four watts, you'll start to see copper 10GbE take off," mostly in Cat6e implementations.
Thomas Scheibe, the director of product management for Cisco's Campus Switching Systems Technology Group, agrees with HP's Hinton that improving power consumption is also driving the move to 10GbE. "The expected arrival of smaller-size and lower-power 10GBase-T PHYs [physical-layer devices] and the upcoming EEE (Energy Efficient Ethernet) standard will accelerate that integration," Scheibe said.
Today, 10GbE switch sales are rising, per-port costs are dropping, LOMs are beginning to appear on server motherboards, power requirements are being reduced, and major players are offering lines of affordable top-of-rack switches.
Intel's Jason Waxman may be right. It appears that 2009 may very well be the year that 10-gigabit Ethernet goes mainstream. ®
Fast(100Mb) ethernet is all most routers need - you really think that manufacturers are going to spring for more expensive tech when home internet speeds are still a pittance compared to the size of the current ethernet pipe? Most datacenters could probably get away with fast ethernet as well; if the router's WAN port is gigabit, even more can get away with LAN fast ethernet. Don't be surprised that the lower end hasn't moved - it won't until it has to(which is why IPv6 won't become popular for a LONG(technologically speaking) time).
I would like to see most routers supporting 1gibit. It seems that unless you know what you are doing you are likely to end up buying a router with only 100mbit ports. They should have stopped selling a couple of years ago.
Lack of single standard delays adoption
Its the same story as with Blue-Ray and HD-DVD. Those who will invest first may loose. Now its the same between RJ45 and SFP+.
Personally I am not sure shall I do the optic or copper for the new rack installations. 10GBASE-T power savings may be achieved - or - may not be achieved, who knows if, and, when. Why twisted pair cable is so desirable? Who really wants to change all the existing cabling to 6E? Existing Cat5e cables won't work, 6e installations even today are very rare, most cables sold are 5e for 1Gbps, or CAT6 and that is not good enough for 10G.
If new cables are needed anyway, do you prefer new optic or new copper cabling? Optic cable is already cheapier than CAT6e cable, and looks more future proof. So why not optic, in the future it will allow 100G (1G some time ago looked as fantastic as 100G looks today).
What works and looks good today is the SFP+ standard, with already available LRM optical transmitters for "datacenter" multimode cable (up to 220 meters), or LR for "telecom" single mode cable (up to 10km).
Maybe the solution is for the servers to provide standard SFP+ sockets on motherboard, with "factory default" plugged-in 1Gbps SFP+ with RJ45 connector for 10/100/1000 connections. Those who need 10Gbps can buy 10G SFP+, either for copper direct connect up to 7 meters for in-rack connection or optic SFP+ if longer distance (out of rack) is needed.
4W per 10GBASE-T port is too much. Also there is no switches with 10G RJ45, and also there is no CAT6e cables in the data centers. For top-of-rack switches SFP+ are needed, not RJ45. If they build a 10G chip that works over RJ45 and transmits only the minimum power needed to work over specific distance, then power usage would be small to the top of rack switches, due to the short distance, and more power needed only if line is connected to the far away switching cabinet directly (up to 100 meters). That may be acceptable, but its not yet developed.
It looks strange when switches go for SFP+, and network adapter vendors talk about the plans to develop RJ45 solutions.
ONE standard is needed, and until that is not happening, I am not going to invest.
The standard choices are either SFP+, that looks good and works now, but is still expensive, or RJ45 with low power, and that is not yet developed. Unless SFP+ sockets on motherboards will start to appear, and SFP+ optic prices will drop considerably in the nearest 3 months, nothing on mass scale will happen in 2009. The sharp and fast price drop for SFP+ modules (at least for 10/100/1000 RJ45 module), and SFP+ sockets in server motherboards seems unlikely today. To accelerate 10G adoption today, server industry needs to eliminate RJ45 and go for SFP+ , or we will continue to wait not clear how long, until power savings will be achieved for 10G over RJ45, if ever.
So, 2009 will NOT be a year of mass 10G adoption. Maybe 2011.