Ethernet will continue to surprise us, with a 40Gbps version likely, and an unexpected appearance of 10Gbps on copper cables, according to Bobby Johnson, CEO of Foundry Networks. Meanwhile, Gigabit and 10Gbps Ethernet are finding uses in the enterprise -- which is just as well for Foundry, since the company has specialised in Gigabit Ethernet, and found the majority of its market in dot-com service providers -- until recently.
"Ten Gigabit is already running on copper -- for distances of around 15 ft," said Johnson. "We need this to be 100 metres, but there should be products that do this in around two years."
In fact there are two separate efforts to put 10Gbps Ethernet on copper cabling -- even though when the standard was first finalised, it was widely believed that a worthwhile transmission distance would never be achieved on copper cables. A November meeting of the IEEE 802.3 group which co-ordinates Ethernet standards, will help determine which becomes a potential standard.
"The first is being called 10GBase-CX4, the second 10GBase-T," according to Intel Fellow Bob Grow, chair of the IEEE 802.3 working group, a principal architect at Intel and a former chair of the 10 Gigabit Ethernet Alliance. The 10GBase-CX4 suggestion uses XAUI (10 Gigabit Attachment Unit Interface, pronounced "Zowie"), which is defined in IEEE Std 802.3ae-2002.
As with other Ethernet standards, XAUI re-uses physical interconnections designed for other networks, in this case 10Gbps Infiniband and Fibre Channel, which can be run on short copper cables. "10GBASE-CX4 is targeted to rack and stack interconnection within an equipment room, by running XAUI over 4X Infiniband cables and connectors," explained Grow. This will require some additional specification to the 802.3ae; standards work on this could be kicked off in November and would be completed quite quickly, said Grow. Among the companies working in this area are Xilinx.
Meanwhile, 10GBase-T refers to efforts to get 10Gbps running on conventional twisted pair cables. "This effort would take more time to generate a standard than 10GBASE-CX4," said Grow. A standards effort for 10Gbase-T might run it over 100m lengths of "horizontal cabling" (specified in TIA 568 or ISO/IEC 11801), but might settle for a shorter distance.
Before a standards effort is started, the proponents have to convince the IEEE802.3 that the effort is worthwhile.
After 10Gbps, the next stop will probably be 40Gbps, said Johnson. Discussions are underway and a standard could be produced within about two years -- with some vendors making products a little before this.
The jump to 40Gbps would be a departure from Ethernet's traditional order-of-magnitude jumps which have taken it through Fast (100Mbit/s), Gigabit and 10Gbps versions. While 100Gbps would be possible, it would be pointless and difficult, because there are no existing physical interface standards at that speed, said Johnson.
The Ethernet community has traditionally adopted existing physical interconnection standards: Gigabit Ethernet used the physical layer defined for Fibre Channel, and 10Gigabit Ethernet used the OC-192 fibre interface. The next optical specification in the telecoms world is the 40Gbps SONET specification OC-768. This is currently the fastest SONET specification, defined by the US T1 committee (and equivalent to STM-256 in Europe's SDH set of specifications).
For Ethernet to jump the gun to 100Gbit/s would require a new physical layer optical specification. This would effectively stake out Ethernet's claim to pre-eminence in telecoms. While it may be that Ethernet is the future of telecoms, the urgency has gone out of the move to set new standards, as the heat has gone out of the telecoms infrastructure market.
Nevetheless, a jump to only four times the speed of a previous version is a break with Ethernet tradition: "Historically, 802.3 as been reluctant to do a 4x version," said Grow. Quadruple speed versions were proposed -- and dismissed -- as the next Ethernet standard, after both the Fast Ethernet and Gigabit Ethernet standards, he points out. "But 10GbE is the first time we are at the same rate as the SONET world. Link aggregation is now the standard method for accomplishing 2x or 4x the data rate."
10Gig Ethernet is going well
While some of us might wonder who could need it, 10Gigabit/s (Gbps) Ethernet is a good market and a growing one, according to Johnson. Foundry launched 10 Gigabit/s (Gbps) Ethernet in 2001 and has had the market for layer 3 10Gbs switches more or less to itself so far -- the sector now makes up 10% of Foundry's revenue.
Much of this market is with service providers: "10Gbps Ethernet is about a fifth the price of OC-192 SONET," Johnson reported. "It's already clear this will be a WAN technology. In the metro space, providers are buying 10Gbps, without bothering with the optical technology underneath." It will take time, he says, before they have enough customers to need the extra bandwidth that optical technology like desnse wave division multiplexing (DWDM) can provide.
More surprisingly, 10Gbps is also apparently required at enterprises: "It's user driven, by people with a perceived need for 10Gbps -- ISPs, exchanges, hospitals, military bases and film houses," said Johnson. And there will be even more take-up in 2003, he said, for such things as government-funded distributed supercomputers.Surprisingly, for a company whose business model was predicated on shipping network kit to the dot-com economy, Foundry has weathered the storm pretty well. The Way Johnson tells it, a massive 50% growth in the company's enterprise business has almost compensated for the disintegration of its service provider revenue, letting the company actually stay profitable. "Since the end of 2000, we have felt the pinch." said Johnson. "At that point we had 65% of our revenue from service providers, and 35% from the enterprise. Today, 75% to 80% of our revenue comes from the enterprise. We still have large service providers, such as AOL and the metro networks, but there is no web hosting business left." The enterprise growth has been helped by FastIron, a bunch of products tailored to the enterprise, launched at the start of 2002. These are transparently Foundry's service provider products, cut down to fit into company networks. "They are cost-reduced versions, but have the same fabric and the same software" said Johnson. For example, the FastIron 1500 costs 35% less than the service-provider version, the BigIron 15000. The difference is in the scalability, the number of routers it can support and the memory. It should be good enough for an enterprise, but for a service provider, with a high number of customers to use it "would be dangerous," said Johnson. If enterprises need to move up a scale, they can plug in BigIron modules, but service providers can't cut corners by trying to get by with FastIron blades. Enterprise themes The main themes in the enterprise are integrated voice over IP, security features, and what Johnson calls "webification." By this, he means the use of higher-level Layer 4 to 7 functions, for load balancing and streaming. These were originally intended for service providers, but it turns out that enterprises have a use for them -- if nothing else because the service providers are no longer there to provide them for enterprises that wanted them. You could sum the benefits up as tidying up racks, agrees Johnson. "Any market goes through a period of consolidation," he said. "In the short term, the economic environment will stifle investment and racks will get clearer. There's a focus on meat and potatoes, and no budget for dessert." The security pitch is based on Foundry's ability to look deeply into network traffic, picking up information from the higher layers of the network model. "We are the leader in deployment of deep packet scan technology," said Johnson. Among other things, this can shrug off denial of service attacks, where attackers swamp servers with multiple spurious requests, because at one level or another it will be obvious that they are not all genuine individual requests. The switches can also load-balance between firewalls and other security devices. ZDNet expressed surprise that VoIP was such a big part of Foundry's plans, as we have yet to see much of it: "A couple of years ago, I was cynical myself," said Johnson. "But integration is real. People do want that cost-saving. We are doing it; we have that religion." Of course at around 10k per voice channel, VoIP is not going to sell many extra 10Gig links -- each one could hold a million voices. But green-field sites, are increasingly installing integrated networks from the start, said Johnson. One such is the Greater London Authority building, set up next to Tower Bridge by Mayor Ken Livingstone -- although in fact, Foundry's partner here, Mitel, has put in traditional PBXs which will be replaced in due course. Migration and partnership is a better strategy, said Johnson: "Cisco would make you rip out your PBXs." The other factor that may eventually drive users to VoIP, is that the evolution of PBXs may simply draw to a close, with companies like Mitel moving across to integrated networks: "Even the PBX manufacturers aren't building the next generation of PBXs," said Johnson.