Bypassing the Stoplights
on the Information Highway

CJ talks with Jeff Shapard of PSINet and Vince Gebes of PSI Japan

interviewed by R.A. Lemos

Before his return to the US about two years ago, Jeff Shapard had long been a leader in Tokyo's fledgling public networking community and sysop of TWICS, one of Japan's first electronic networking systems. He is now a senior product manager with PSINet.

Vincent Gebes is currently assistant general manager for PSI Japan.

Jeff, you're best know here for setting up TWICS and helping to bring the Internet to Japan. What are you
doing now?

Shapard: I was involved in online services when I was here in Japan; that was essentially pre-Internet, although, I did have a small role in setting up some of the early commercial Internet services. My role now in the US with PSI is Senior Product Manager, Corporate Services Marketing. My focus is on the InterFrame service, which in the US is our core product -- our dedicated circuit service, 56K to T1 access.

Dial-up services used to be 2400 bps; I remember when 1200 bps was high speed! But now, for all organizational connections, your corporate needs, unless it's a pretty small office, you don't use dial-up connections -- except ISDN, which has become very popular. Japan has been very progressive in spreading ISDN; NTT loves ISDN and has put a lot of money into building it up here in Japan.

ISDN is not so widespread in the US; it really depends on the region, with each of the different regions in the US being operated differently. Some regions are quite digital. It's actually widespread with flat-rate in some regions; others don't even know how to spell it yet.

Do you foresee any problems with ISDN?

Shapard: Problems with ISDN? I think ISDN is a wonderful thing. It is a way to take this very large and expensive copper-wire-based telephone plant and get a lot more out of it, give it more life. ISDN is great.

Will we soon need to go beyond ISDN? Will 64K or 128K be enough?

Shapard: I don't know how much farther we'll be able to go with copper; 64K or 128K is a lot of bandwidth. I think what we'll see is much more efficient use of bandwidth. We'll get more compression and the emergence of better protocols -- such as multilink PPP, which essentially allows you to bond multiple B-channels together to create more bandwidth. Typically, what people do now is use a 64K channel or link two together for 128K. With multilinking, you can add an infinite number of B-channels. If you have lots of ISDN connections, it spreads the individual links across all of these connections.

What's the average bandwidth on the Internet?

Shapard: We try to talk about aggregate bandwidth -- how much circuitry you have, and how you route your packets. If you have T3 lines between a bunch of routers, it's like having stoplights every three blocks: It doesn't matter how fast it goes between stoplights, you get these points of congestion along the way. And you end up with a lot of hops. The more hops, the greater the congestion -- and the greater the congestion, the lower the throughput. We designed our network to minimize the number of hops. PSI at last count had 40,000 miles of network.

What's the alternative to using routers?

Shapard: A traditional protocol network is based on routers. All providers in Japan are currently doing it this way. What they do is establish a circuit, from point A to point B, then put a router on either end, and -- bam! -- they are an Internet provider. If they want to expand to another region, they get a circuit and put a router up. Another circuit, another router. So they end up with this big web of routers and circuits, which is inefficient.

Routers are really great devices; they are smart, they are robust, they are designed to handle the sort of situation where part of the network goes down and they have to figure out where your packets need to go by routing them in a different direction.

Any serious Internet provider in the US has T3s -- 45M-bit lines. Typically, they have it across the busiest stretch of their network, and the rest is T1s. So they can say they are a "T3 network." Yet, the problem is that packets must make it from an outlying area (sometimes making four or five hops) and move across to the main backbone; invariably there is congestion. So even though there is a T3, the routers [on either end] are not going to be able to maximize the throughput. You get a lot of hops, a lot of congestion. For instance, in a busy metropolitan area you might just have one big router with all kinds of circuits coming into it; that's not efficient. It's OK for little networks, and all right for a company's internal networks, but....

We set up our network so that it is a fabric, a mesh of circuits. And instead of being physical circuits across the network, we use frame relay technology to set up a network based on what we call a layer-switching fabric.

Do you think it will take more time for the Internet to launch here in Japan?

Shapard: There is always the tendency in Japan to talk about how far Japan is behind the US. But,it is more like how far the US is in front of everybody else. The US is where so many new technologies emerge, and are rushed off into market and evolve while the rest of the world is just sort of watching to see what works and doesn't work. Japan is in the midst of a slingshot effect, where everyone feels that "we are falling behind," and so the tension builds, and suddenly it leaps forward and catches up. Japan will leapfrog some of the technologies that America has gone through.

LANs are a good example. You see many American companies that have layers of legacy LAN networks in the same office building with no organization. In Japan, however, I see the whole company on one kind of LAN. When it grows old, they wait until they see where they want to be; they let that stuff get old and creaky, and then dump the entire thing.

I remember, in the '80s, Japanese companies had a couple of processors up against the wall and a maybe a few PCs in accounting, but everything else was done by hand. And they saw it coming. They had a big investment in the mainframe back in MIS and the data processing department, so they watched and waited, thought about what they wanted to do and explored their business processes instead of just adding PCs into processes (as happened in the US so often -- inefficient processes were computerized). They thought about what they wanted to do and how to do it most effectively, redesigned their processes and decided on their environment, and then -- bam! -- in six months, there was a notebook computer on every desk, with a big fat fiber connector to their LAN, all tied into their distributed computing environment in the back room. A lot of thought was put into the network itself.

Gebes: In regards to Internet technology in Japan, I see Japanese Internet providers moving forward very quickly -- in modem speeds, for instance. We recently upgraded our modems to 28.8K, yet many of the Japanese providers started out with 28.8K modems. And they are taking advantage of the new technologies that are rolling out, rather than basing their network on older base technology and then upgrading. This is good; it's helping build up the Internet here in Japan.

Another example is the widespread use of ISDN in Japan. It is amazing. Only in the last year or so has ISDN picked up in the US. The pattern in some places in the US was that telephone companies started offering ISDN, and some businesses said, "I want that 64K access." So companies started coming out with 64K equipment. In Japan, though, because the ISDN network was deployed so quickly, users didn't have an application for it at first. There was no great demand saying, "I want 64K access," so what happened was many of the Japanese manufacturers started coming out with TAs (terminal adapters) which will connect a regular phone or a PC through a serial port to the ISDN network -- originally at 19.2K, not much faster than a modem, then at 38.4K. Finally, because of the Internet, you're starting to see people saying, "I want 64K."

PSI Japan supported 64K access from the start. However, there was no 64K terminal equipment, so we have had to downgrade our systems to offer 38.4K access to our users.

So, does this mean that Internet development in Japan will parallel telephone development?

Shapard: The Internet is based on shared data space -- that is, packet-based networking. You have all these folks with different applications sharing a different number of circuits. When you start getting into applications like video phones and MBone and CU-SeeMe, and these transmission technologies, you really need to have a lot of bandwidth allocated to that particular stream; if you start to break it up, it gets really choppy. So it's more of a switched-network technology rather than a packet-network technology.

Gebes: One of the other things you need to consider is that, although the technologies themselves are converging, it is all one physical layer. Historically, countries and territories have regulated their telephone companies very strongly -- and their broadcasting companies. So there is a question of whether those things can ever integrate. Where Internet providers and telephone companies will integrate is very much a function of the regulations of a particular country. And this is something that is still evolving in Japan.

Shapard: Remember, the Internet is a network of computers. When you start integrating telephones and videophones, that's a different kind of animal. It has to do more to do with the digitization of these other technologies than it does the Internet spreading out to take over these other technologies.