MINNEAPOLIS - Few organizations can match the Internet Engineering Task Force for healthy debate. Every proposed protocol seems to bring up four or five different implementations - and possibly double the opinions to back them.
Network World Senior Online Reporter Sandra Gittlen yesterday met in Minneapolis with IETF veterans Fred Baker and Paul Ferguson to discuss the goings-on at the 44th IETF. Baker is currently IETF chairman, as well as an engineer at Cisco, specializing in network architectures. Ferguson has authored several standards and he also works with Baker at Cisco, another network-architecture specialist. But they have different views when it comes to the future of the Internet. Ping Pan, a researcher at Bell Labs, joined in for the lively debate, which covered everything from voice services to videoconferencing to policy-based networking.
Ferguson: Policy's a splintered effort here at the IETF.
Baker: Well, we're trying to solve an insolvable problem. But in the end, there's going to be a rules-based expert system in every router.
Ferguson: What does the policy server look like? It could be housed with a [Lightweight Directory Access Protocol] server.
Baker: It could. But you definitely want a cached instance of LDAP with it.
Ferguson: I'd love to see LDAP directly linked to policy.
Baker: The problem is that LDAP basically asks: "Have you changed your mind about anything yet?" There could be info in three databases and only when the database is asked for an update [with LDAP] does it give one. Couldn't they have figured out publish/subscribe by Version 3?
Pan: Policy could use [the Common Open Policy Service], it's a lightweight protocol. Of course, security is hampered because it's light, but it's okay as long as policy is not outside of the domain.
Ferguson: ISPs don't want an interdomain protocol. Look at B-ICI [B-ISDN Inter Carrier Interface] for ATM. It was a fully defined spec for how to exchange information between ATM nets. It didn't go anywhere. Telcos don't want other telcos to know what their internal topography looks like. The telephony mentality and packet mentality just don't synch up.
Baker: Yeah, it does, it's called ATM.
Ferguson: People in telco industry think that they can translate across both networks.
Pan: They want to reuse existing protocols like Signaling System 7, but with the cost of adapting, it might be easier just to develop new ones.
Ferguson: Remember, there's the Internet and there's other IP networks. If you own a piece of the Internet, you better be concerned about connectivity. You better make sure people can get to your Web site and click on a page. It's all about economics.
Gittlen: So when you say connectivity, you mean make sure you have enough bandwidth? Just keep boosting it?
Ferguson: Yeah.
Baker: So throw more roads at Silicon Valley and the traffic problem will disappear?
Ferguson: There you're talking about real estate. There's no real estate on the 'Net.
Baker: But look what happens to the phone system on Mother's Day. No amount of bandwidth can handle it.
Pan: That's why we're building classes - people are willing to pay more money to guarantee they can get what they want. A billionaire will pay 1,000 times more to get the quicker click.
Gittlen: Then the protocols are just creating a service market for ISPs?
Ferguson: No, it creates more efficient protocols.
Pan: You keep adding bandwidth and people will use it up.
Baker: If you can know how much traffic and how it will be treated, then you have predictable traffic. There are three things we know about the Internet: fairness, conditioning and shit happens packets get lost. So I say to you, you pay me so much and I'll give you this much bandwidth. That's fine. But suppose we put in voice. Taiwan's doing it right now. Suddenly, you have fairness with First In/First out, you have conditioning. But when you get to shit happens, voice won't accept this. That's why you have Integrated Services to take care of that.
Ferguson: Voice riding on the 'Net could mean the end of the central office. The network will be packet end to end.
Baker: Cisco sells packet telephones now.
Ferguson: Voice only works on a system where the network is engineered not to lose packets. DSL is a cool service. Yet RBOCs and CLECs are not tripping over themselves to deploy it. Why? It might cut off their voice business.
Baker: No, they're not jumping on the bandwagon because of the backbone. It needs to be beefed up and they're not ready yet.
Pan: Remember, ATM was supposed to rule the world. Are we looking at the same thing with voice over IP? Is it too hyped?
Baker: Could very well be.
Ferguson: I don't think so. There are huge economic drivers for voice. But that doesn't mean there's a solution. And remember, there were no applications for ATM.
Baker: 75% of the Internet's traffic hits an ATM network.
Gittlen: For the quality of service?
Baker: ISPs are using unspecified bit rate, not quality of service.
Ferguson: They're shipping packets across the network as fast as they can. They prefer to do packet shaping at the network layer.
Baker: ATM gets them on fiber. That's why they like it. The line is more expensive, but the amount of bits carried more than make up for the cost.
Gittlen: What about the ITU's H.323 vs. the IETF's Multiparty Multimedia Session Control (MMUSIC) videoconferencing standards?
Baker: They're different applications. MMUSIC is a meet-me protocol, which is fine for broadcast TV. But if you want to place a phone call across the 'Net, meet-me doesn't work.
Pan: Session Initiation Protocol in MMUSIC works better than H.323 because MMUSIC is Internet native.
Baker: I'll tell you the biggest thing with H.323. Take a look at the title on the H.323 document. The headline is "Videoconferencing over a local area network." What else can I say?
RELATED LINKS
IETF working group seeks to improve security alerting
Network World Fusion, 3/16/99
