A life in the fast lane

Bernard Daines, CEO of Gigabit specialist Packet Engines, has been in the industry for about two decades and has been instrumental in a number of developments that we now take for granted.

Paul Grant talks to the man dubbed the Ethernet king, and gets an insight into the world of Gigabit Ethernet from one of the industry's most influential figures.

How it all started

I've built things for other people ever since I was in Junior High School, and through college, in fits and starts. I worked for Hewlett-Packard and IBM during and shortly after my college days, designing computer circuits, doing printed circuit layout programs.

In 1977 I set up and ran a development company, until 1992. In that period of time, I did 3Com's first controllers in the early 1980s, and in about 1988 the first 10BaseT switch that was ever built - for a telecom company.

I did a lot of other products over that period of time for a lot of companies and in 1992, of course, I started Grand Junction Networks. I left there in late 1993, re-constituted my own company and changed the name to Packet Engines. I went from contract developing almost anything to concentrating on IP licensing in the Ethernet area. We began by licensing 100Mb Macs and then moved on.

In late 1995, we had the idea of extending the same concept when we came up with 100Mb Ethernet. The thinking behind this was to take a fire circuit that already existed for FDDI copper, marry it to a speeded-up Mac and we had a no-technical-risk 100Mb solution. We noticed in late 1995 that you could buy optical devices for Fibre Channel and we had the notion of doing the same thing again. We took funding into Packet Engines in June of 1996 to build the big switch, which is still the highest performance switch in terms of non-blocking.

Getting into Gigabit Ethernet

The amount of traffic is continually growing and so is the number of users, and I don't think it shows any sign of stopping any time soon.

The internet is doubling every hundred days or so, files are growing, there are more realtime applications, as well as voice and video. We're moving into the Gigabit Metropolitan Area Network (Man) area, where we are tying buildings together across the country. Normal people - up to now - haven't been able to afford that because T1 (E1) lines and such are very expensive. But by linking buildings together with Gigabit Ethernet, that suddenly becomes very, very practical to do. We see the demand for these kinds of services continuing to rise, and that's the direction we have taken.

In October 1995 we held an industry meeting where we had 30 companies and 50 people from all the names that kicked off the Gigabit Ethernet operation. We started the process of working towards IEEE standards in November. The only trouble was getting everyone to agree once the first concept got out. It went around for a while and circled back to exactly where we started, with just a few little details left to work out.

When we took a look at the Fibre Channel stuff, it was at 1.025 Gigabaud.

That would have been 800Mb Ethernet, and that didn't seem to have a very good marketing ring to it. We went off to the component vendors and said "What can you do about kicking this up 25 per cent?" The optical people said there was no difference between 1Gb and 2.5Gb, that was easy for them to do.

Those making the serialiser/deserialiser had already been sampling their circuits up at 1.25 Gigabaud, so that wasn't an issue. But there were two main complications. Firstly, part of the committee looking at the optical stuff determined that everyone's parts would survive in the standard, so it took three months to converge on some testing issues. Secondly, I tried very hard to get half-duplex thrown out, because I didn't believe that half-duplex at Gigabit would ever be used. Others said if it isn't CSMA/CD it can't be Ethernet. That's gone now, full-duplex and link flow control are full standards. Gigabit Ethernet is a standard, and there isn't anyone I know that is building half-duplex Gigabit Ethernet.

We have a hub that is a full duplex hub, which uses the X standard for both flow control and full-duplex. It's far superior to any collision contention issue. If you think about it, ATM would never have the stronghold that it has if Ethernet had been thought of as a predictable system, and not a collision system. So, in fact, by insisting that Gigabit Ethernet carry half-duplex, we delayed the conversion of all the effort into qualities, predictabilities, flaw management and all that. I can guarantee you that there will be no half-duplex in 10 Gigabit Ethernet. Unfortunately I was unable to persuade enough people to get rid of half-duplex, there were too many purists there. In hindsight they wished they had. A lot of effort went into making that standard which no-one will ever use.

Metropolitan Area Networks

There is a drive for bandwidth from organisations that have several buildings, and connecting them is an issue. 100Mb Ethernet with copper is limited to 100m, with FX you can go a little bit further, but it doesn't provide much extra bandwidth.

In my mind, it's been an issue of Gigabit Ethernet as a backbone solution running between buildings. Think of companies like Microsoft, with their big campuses and multiple buildings in various places - it's natural to want to be able to connect them together. So the idea of doing that was always there.

We also had opportunities to extend the distances further. The standard says that long wavelength laser on fibre is only standardised to 5km, that's because there's a lot of fibre out there that wasn't built to its own specs, or current specs. But if you have good quality, relatively new fibre and good quality optics, with just the standard LX laser you can get 20-40km easily. By putting an LH laser in you can get up to 100km.

We have been testing that, with the set-up in Stockholm and actually demonstrated 60km on fibre that was being laid. We're running raw Gigabit Ethernet direct through there, because as long as they stay with links that distance, we don't have to have repeaters out in the infrastructure.

This is definitely going to be a huge area for Gigabit Ethernet. If you think about pushing Gigabit Ethernet into Wavelength Division Multiplexing (WDM), Dense Wavelength Division Multiplexing (DWDM) or POS, you can just take the same framing and use the extra structure that carries a longer distance.

You've also got a movement to IP. Our boxes support IPX, IPv4 and IPv6 and eight other protocols, in a combination of hardware, to protect the customer's past and future investments. Then you've got these high performance, wire speed, Layer 2, Layer 3, Layer 4 routing switches at a very low cost compared to Cisco boxes, and you have the availability of dark fibre.

This allows people to build networks they never dreamed they could before.

A lot of this is easy to do and is suddenly possible at a low cost.

The Death of ATM

I remember the days when, every year, some of the magazines would say that this year was the death of something or other. It's the death of core memories they used to say - that happened every year for about eight to nine years - and then they quit saying it and semi-conductor memory was king. It took a long time for core to get displaced in the general marketplace.

ATM isn't going to die overnight, because there are people who are committed to installing it. But with the capabilities moving into Gigabit Ethernet equipment, of having all of the services and more so than ATM, there really isn't any reason for ATM to exist. I believe it will be supplanted over time.

Gigabit Ethernet to the desktop

At the moment Gigabit Ethernet to the desktop goes to those people who want absolute performance. For instance, in our software lab, we have every one of our software engineers with a Gigabit NIC in their machines.

This is tied up through one of our hubs to shared switch ports, into our switches and then into our servers, because when they do half a million lines of code and archiving, there's a lot of traffic. So they're able to use it in the thirsty environment that the high-speed user may want to get the benefit from.

There aren't very many applications that use Gigabit very fast, but there are enough that people want to go above 100Mb. The fact that they can't go up to 1,000Mb in today's NT machines doesn't matter. The fact is they can go faster than they could before. The volume in that area won't kick off until the copper Gigabit Ethernet comes along and prices come down.

We were really smart folks when we started Grand Junction Networks, we did a lot of stuff. I remember going out and telling people, when we had the first 100Mb NIC card for about $1,500, "Wait, be patient, we'll get it down to $295-$250 some day and it'll be fine". Look how wrong I was, you can now buy 100Mb cards for $39. But, in fact, by 1999-2000, the cost of the number of transistors to do copper Gigabit Ethernet will be no different from the cost of the number of transistors to do 100Mb back in 1994. I fully expect that you will be able to buy Gigabit Ethernet copper NIC cards in the $200-$250 area before too long, and it probably won't be too long again before they get down to $100. But they don't need to to be compelling right away, and people will start doing the same thing that happened with 10/100Mb.

Nowadays, if you are a corporate purchaser of PCs buying a networking card, you are out of your mind if you don't buy 100Mb. Even if you can only use up to 10Mb, it is minimal money, and it will cost you more just to open the box to change it from 10 to 100 later. So when the 100/1000Mb card on RJ45 copper comes out shortly at around $200, even though its $150 more than a 10/100 card, that $150 is peanuts compared to changing it a couple of years later.

There will be Ethernet traffic carried on Terabit networks, but most of these Terabit networks will be WDM multiplexing, not one Terabit stream if you wish. I think that the proliferation of a Terabit stream, one individual Ethernet stream at a Terabit, isn't terribly likely to be a big deal right away. I see Gigabit Ethernet, two and a half Gigabit Ethernet, 10 Gigabit Ethernet being carried on these Terabit networks, and combinations of them giving you traffic that goes up above Gigabit, I don't see it right away as individual streams of Terabit.

The future of Ethernet

Ethernet has moved from being a peer system from the early days, to a system that went with 10BaseT into a starwired system with some management and supervision. It was still collision at that time - even though the technology was full-duplex - but we've moved on now. Ethernet now is really the framing, the packet. It is carried by multiple technologies, bit codings, speeds, and ways of handling it in the raw form. It has come down to Ethernet being the framing and the management, it doesn't mean half-duplex any more.

There's no end to the future there, if you look at it, over 85-90 per cent of the nodes of networking that are being shipped in the world today are Ethernet. There is just no stopping it, and no reason to stop, because the beauty of it is you can buy a box, with new capabilities and speeds, hook it right into your network, and it's all compatible because the frame is there. Nobody has to chop it up or translate it. It's very comfortable for the users to be able to do that.

We are going to continue to add interfaces, we've got FDDI and FX, we've got ATM and packet over Sonnet coming, we're putting all kinds of things into our box. We're finding that this box is not only going into the backbone but also going into the ISP areas, as kind of a crossover.

You'll see an announcement coming out of London soon, along the lines of what's happening in Stockholm (see news). We're going to do something similar in Poland, and we've just finished working in China - using the fibre that they use for the broadcast and cable TV industry. The future of Gigabit Ethernet and Packet Engines is very bright in that area as well as the traditional backbones. Then we'll move on to bring out bigger and cheaper boxes. We chose to bring out the big box first instead of getting into the pennies-per-bit war that others have succumbed to, and which will have a high casualty rate.