For several years, Judy Estrin has been an outspoken critic of what she calls the "Band-Aid" approach to strengthening the infrastructure of the Internet. The former Cisco CTO and her husband, Bill Carrico, also from Cisco, launched PacketDesign last June as their latest contribution to making the Internet work better. The company has raised $24 million in funding and announced its first spin-off, Vernier Networks, to make wireless Ethernet usable by businesses. [email protected] Partner editors Deborah Gage and Eric Carr caught up with Estrin at PacketDesigns headquarters in Mountain View, Calif. She offered plenty of insight into her company, the economy and the Internets evolution. Heres a sampling of her thinking.
SP: What projects is PacketDesign working on?
Estrin: Were trying hard not to be a gun for hire. We want to hear about peoples problems but solve them only if theyre generally applicable. We identify projects—there are some ideas here. We can do projects for years based on [chief scientist] Van Jacobsons ideas, and some come from meeting with people and talking about problems. Were aggressive about managing what we do and what we dont do, and about stopping projects if the hypothesis not correct.
I really enjoy the building, and I dont believe in incubators, although there is a place for incubation as part of something else. But the problem with incubators is the end goal: turning out companies, so youre rewarded on the number press of releases you wrote. [Also,] most of the work on the Internet today is coming from industry. There used to be a lot of good work done in government labs, and thats less now. And theres a brain drain—people go where they get equity. Stanford set up a networking and wireless research center with industry; people are trying to deal with this problem, but theres a gap.
For PacketDesign we purposely picked projects that spanned a lot of areas—some will come to fruition right away. Our first spinout is in the wireless space, and other things were doing will span everything from security areas, to a couple of things in the routing area, to projects in the network system interface area.
SP: Has the market downturn affected PacketDesign?
Estrin: Cycles exist in the stock market and always have existed. I think the downturn is not a V [a steep decline followed by a rapid recovery] but more like a U [an extended bottoming out followed by a recovery]. Even after recovering, were not going back to where we were over the last couple of years. The dynamics of the market is not just about interest rates. If you look at what drove growth, you had a new product phenomenon: PCs, cell phones, the Internet, a number of things driving that growth, and its just sitting today. You couple that with interest rates and oil prices and energy.
In hard times, people have even more of a bunker mentality, and we have to fight against that and not use the economy as an excuse not to do the right things.
SP: Has the notion of "Internet time" helped or hurt the Nets development?
Estrin: Speed is good, and Internet time for the most part is good, but its not an excuse not to think. When time to market is so critical to competitiveness, you get a culture where engineers dont think through the implications of what theyre doing—they solve todays problem today and tomorrows problem tomorrow. Theyve got a finger in the dike. PacketDesign looks a little longer term—theres nothing wrong with Band-Aids, but we want you to get stitches if you need them.
The Internet now has different types of traffic and not just more. [The traffic is] put on a network not originally designed to anticipate that, so how do we deal with service for new types of apps? Because fundamentally, Internet Protocol and routing are the right way to go; you can add in things like quality of service done the right way and other things in the routing area, so you do not have to go to an alternative architecture. Theres a disadvantage with switching—a big issue with switched architecture versus routing has to do with scalability.
The beauty of IP is that each packet carries a globally known address, and each node can make its own decisions. When you start putting in state, when you take away the fact of globally known addresses, you take away flexibility. It often happens in the industry where an architectural vacuum comes up. MPLS, for example, was designed to solve the mapping of ATM and IP together, and its not a bad solution for that specific problem. But it should not be used end-to-end as a network architecture. Its gotten out of hand.
SP: Can IP version 6 correct some of these problems?
Estrin: One of the very important aspects of the design of IP is separation of the control plane from the forwarding path, where the data packets come in and the routers deciding how theyre going to go out. The control plane is the way routers interact to build tables so the forwarding path knows how to forward packets, and IPv6 impacts the forwarding path. There are still lots of things that need to be done here. No one does anything until the customer asks for it, and the customer asks for it when the tidal wave is upon them and people scurry around.
SP: What challenges do you see for voice and data convergence?
Estrin: Quality of service needs to be provided, and one of the big problems today is we have people coming from the telephony world who dont understand IP-based networks. Part is education and part is tools and figuring out how to manage an IP-based network by people who have been used to managing wires.
The networking business has never been aggressive enough with management tools. Multicast was so slow in the beginning because people couldnt diagnose problems—no one had invested in the tools alongside it. Network management has historically lagged behind in the IT world, where in telephony a huge amount of money has been invested in OSS systems and the support systems that go around it.
SP: Should the Internet be regulated?
Estrin: The beauty of the Net is that you have an interconnected network with no central intelligence, so you can add here and peer to here and access the whole thing. Thats why the Internet has grown the way it has, because theres no central authority to set up the connections, and I would not take that away. But people have to think about that—if you are a customer, you should have redundant paths to the Internet. If you have a huge mainframe system, you dont put two plugs into the same circuit breaker. As customers start asking questions of ISPs and get smarter, they do a trace route to see where their traffic is going and know whether theres a vulnerability and either change it or force their vendor to change it. But if you try to have a central authority plan the Net, youll end up taking some of the negative aspects of the telco world and hierarchically limiting the scalability.
Whenever theres explosive growth like weve had, it grows at the expense of some other things, like security. As customers become more dependent on technology, they need to push back on the vendors and ask those questions. But technology has to be there to support it once those questions are asked.
SP: Do proprietary standards, particularly in content delivery networks, threaten the Net?
Estrin: As the content distribution architecture starts to pervade the network, you end up with proprietary technologies. Theres nothing wrong with Akamai, but Akamai is a point solution, and if we start depending on them were locked into that technology. Over time, we need to think about content distribution in a more layered fashion.