The new and improved IP
From Internettelephony on-line.
The new and improved IP
Convergence and the future of the Internet force providers and vendors to look at enhancing IP. But how it will change and what services it will enable remain to be seen
SUSAN BIAGI
The single most ubiquitous network protocol is coming under fire. IP, the transport mechanism underlying the Internet, is outgrowing its original purpose and beginning to show some cracks. The culprit? Success.
Originally developed as part of a communications project called ARPANET for the Department of Defense, IP was created to support an open architecture network that could link multiple, disparate networks via gateways. Vinton Gray Cerf, now senior vice president for Internet architecture and technology at MCI WorldCom, was one of the original developers of the TCP/IP protocol suite with Robert Kahn back in the 1970s.
ARPANET was a success and later was opened to research and educational institutions. That spawned a slew of LANs, which then were connected to what became the Internet. The Internet became commercially viable in the 1980s; today more than 200,000 networks are connected to the Internet, and the number of computers linked to the Internet falls between 60 million and 200 million, by Cerf's estimate. He expects that 900 million Internet-enabled devices will vie for connectivity to the Web by 2006. These will include a host of IP-enabled objects such as wireless phones, pagers, personal digital assistants, TVs, cars, refrigerators and other household items.
Along with the growth of users and devices connected to the Internet, its applications are changing. Today's killer apps are the Web and e-mail, respectively comprising about 75% and 15% of the traffic that flows across the global network, Cerf says. Voice over IP, having overcome its disappointing debut, is making headway, and the ability to transport video over IP will further propel the Internet. The future holds the promise of real-time everything, on-demand everything, inexpensive voice communications and applications never imagined.
IP has emerged as the preferred application delivery method, and equipment manufacturers and application developers have rallied around Cerf's much-touted "IP everywhere" campaign. But we're not there yet. The current version of IP, known simply as IP or IP version 4 (IPv4), faces several technical challenges to enable this dream of IP ubiquity. Security, reliability, service quality, guaranteed service levels and addressing schemes to support the rise in linked devices still are unresolved issues.
No more quality constraints
To combat these and other perceived flaws of IPv4, an industry consortium is working on the next generation of IP, dubbed IPv6. The most notable advancements regard addressing and quality of service (QOS) issues.
On the addressing issue, some fear that the available pool of IPv4 addresses is dwindling rapidly because so many devices are linking to the Internet. In fact, Japanese researcher and programmer Jun-ichiro Hagino expects IPv4 addresses to run out by 2010.
The problem is that IPv4 uses a 32-bit address field. Theoretically this can support up to 4.2 billion devices, Cerf said in a recent article on CNN.com, but he suggests it is unlikely because "the allocation of those addresses has not been very efficient."
The proposed IPv6 uses an address field of 128 bits, which would support 1038 devices--drastically more than the current IP (Figure 1). IPv6 incorporates the encapsulated security payload, or ESP, for encryption and an authentication header--items that do not exist in IPv4--to make transactions more secure. The specification also includes a flow label to support real-time traffic and automated connectivity for plug-and-play use.
A key constraint for IPv4 is its QOS issues, which make it ill-suited for delay-sensitive media such as voice and video traffic. The IP header includes an unused field, which can specify service levels based on DiffServ.
More promising for complex networks are tag-switching solutions such as multiprotocol label switching (MPLS), which are backed by Cisco Systems and other large vendors. MPLS uses a "tag" on the header to identify the origin, destination and path so that the IP packet can flow through the network more quickly. Each device the packet travels through doesn't have to read the header--it simply reads the tag and forwards the packet. The IPv6 spec includes a label, designed to improve IP's ability to support real-time traffic.
IPv4 could benefit from new control protocols, but not changes to the packet format, says Larry Lang, vice president of service provider marketing at Cisco. Coupling control protocols such as MPLS, DiffServ, resource reservation protocol (RSVP) and media gateway control protocol (MGCP) with network planning techniques can make IP viable for real-time transmission such as voice over IP, he says. Those techniques can include limiting bandwidth use, domain switching and traffic engineering.
MPLS is "a step in the right direction," says Michael Arellano, a consultant with Degas Communications Group. "But what happens if you have a bunch of traffic and everything is high priority? Everything can't get through at the same time." Voice and other delay-sensitive traffic generally will remain on circuit-switched networks until that issue is resolved.
Moving forward, looking back
Despite industry support for IPv6, an "Internet upgrade" is no small undertaking. It will take years to upgrade all the IPv4 devices connected to the Internet.
Adopting "IPv6 is a slow process, mostly because IP's biggest strength is that it is installed on millions of computers and routers around the world," says Gunther Gee, vice president of technology at oCen Communications. "You have to do a protocol upgrade to change the protocol standard. That's a slow evolution."
It also means that IPv6 must be backward-compatible with IPv4. "In the world of communications, we can't leave people behind," says Fred Harris, director of network planning and design at Sprint.
IP will have to morph into a slightly different protocol to survive as telecom and datacom networks meld, Harris says. The "new" protocol--whatever it may be--must connect to old systems and devices, including legacy Class 5 switches. It also must support wireless applications, DSL, copper, fiber and wireless links.
"It is clear that protocols are converging, like wave division multiplexing and Sonet are coming together. I anticipate that IP and ATM are coming together," Harris says. IP and ATM have been continually improved "to deal with their deficits. Enhancements will bring the two closer together. It's still going to be called IP, but it will have a lot of characteristics that people associate with ATM," he adds.
ATM offers variable-length cell switching and few or no latency issues. But it is designed to handle high-priority voice traffic via dedicated connections and is kludegy for data transport. The voice support is key, Harris says. "Chief technical officers of start-up [service providers] have looked at all the technologies and are coming back to ATM because of the voice issues. They can use ATM to support IP, but they run into problems when they use IP to support voice," he says.
The fusion of IP and ATM is the most promising evolution of IP. "ATM QOS is complicated," says Ross Callon, formerly chief systems architect at IronBridge Networks and now with Juniper Networks. "Part of the advantage of IP is that you can [move forward] really fast. It's nice to keep things simple."
For at least 10 to 20 years, networks will be ATM and IP hybrids, Arellano says. "IP is just a protocol. If it weren't for latency, reliability and interoperability issues, we could technically switch over today."
Most equipment that supports IP and ATM routes the IP packets and switches the ATM cells. In a few years, that will change, and devices will switch IP and ATM traffic natively.
That introduces another point. "The protocol is not the only thing that needs to change," says Dan Mangelsdorf, vice president of Internet telephony marketing at Nortel Networks. "The hardware and software used for routing IP has to be upgraded, and it has to be upgraded without having to take down the network. That doesn't have anything to do with the protocol itself. We see that as important an element as having reliable service."
Also important is the services aspect. "It's not about the technology, it's about the applications," says William Goers, director of offer architecture for service provider networks at Lucent Technologies. "The voice network is more reliable than the data network. That means we have to figure out a way to make the data network as reliable on the application [side] as the telephony network."
Cultural crossroads
Network origins seem to indicate the IP bent. Traditional telecom companies tend to emphasize reliability; datacom companies focus on flexibility. Changing your position can require a cultural shift in thinking--and it's hard for companies to agree on the issue of IP networking.
"The notion from the telephony side is that if you ever drop a packet, it is a disaster. You have to understand that dropping a packet is no worse than a busy signal," says Cisco's Lang. "When you make a telephone call, quite possibly you'll get a busy signal. That's a probability you engineer for."
IP can handle many data types, including real-time traffic, Lang says. Supporting voice over IP "has more to do with procedures than protocols," he says. With "rigorous" procedures, five 9s reliability can be achieved on an IP network. It's a question of engineering, Lang says.
"You could engineer the network so that only one in a billion packets get lost or it might be more economical to do one in 10 million," Lang says. "In any network, you choose how you want the network to respond. It can reject the session, like the telephone network; it can reject certain packets, like the Internet; or you can use a hybrid, like RSVP on the Internet." The network response is based on the customers' willingness to pay for higher reliability.
Acceptable packet loss depends on the application, and the same holds true for voice, Callon says. "With voice you can drop a certain low percentage of calls. But if you have a busy signal all the time, people will notice." Reliability is a technical challenge, but the level of reliability depends on the application, he adds. "Wireless phones don't have five 9s reliability either, and that is an area where there is rapid growth."
But not everyone is convinced that IP will sustain voice in the mass market.
"It's easy to set up a voice-over-IP network in a small campus or in a LAN where you can control the traffic, but if you have a nationwide network using voice over IP, you have issues with scale and network congestion," Arellano says.
As recently as two years ago, one long-haul IP backbone provider admitted that it could guarantee service because it used only 4% of its network capacity. That's a short-term solution, but as traffic increases, throwing bandwidth at the problem will be inefficient and expensive.
Apps to the masses
But focusing on the underlying technology only covers so much. The true issues are application-driven and customer-centric, says Ben Ho, director of marketing and sales at oCen. A relative newcomer to the communications space, oCen operates a managed IP network targeting the Asian business-to-business market. Its network links Los Angeles, Taiwan and Hong Kong and supports 50 million minutes per month.
Offering business customers integrated services over the IP network requires guaranteed service levels, something the Internet is poorly equipped to handle, Ho says. oCen can control the quality of the information that flows across its network, without resorting to severely limited bandwidth usage, he says.
Although oCen supports the IPv6 efforts, the company is developing its own traffic controls and QOS solutions. "IPv6 will do some great things, but it's not going to solve the puzzle," Gee says. "The key to the future is active admission protocols."
By determining the amount of traffic that can travel on specific network segments, oCen can control the QOS, he says. It requires added gear near the customer and the ability to push voice traffic to the public network when the IP network is overloaded or can't support real-time traffic.
Access equipment manufacturer Sedona Networks also is interested in the progress of IPv6, but the company maintains its edge services development (Figure 2).
"IPv6 will help us tremendously," says Joseph Elchakieh, president and CEO of Sedona. "It addresses good issues, but it focuses on the core. Companies like us are now realizing the access network is a different animal."
In truth, Lang says, there is "nothing IPv6 can do that IPv4 can't do with some bit of jiggling. The incentive is for the network management staff because you don't have to change the end system." IPv6's adoption will weigh the needs of network managers against those of applications and users, he adds. Until applications demand IPv6, there is "no reason for users to tolerate the transition," Lang says.
Still, "IP demand is more fundamental," Lang says. "End users generate IP traffic; they don't generate ATM traffic. The basic demand out there is for IP."
Research from Cahners In-Stat Group backs that, noting that revenues from IP services will increase dramatically by 2004.
"IP is the service delivery platform," says Lucent's Goers. "It doesn't make sense to argue that any longer, and whatever evolution path it follows is going to be the way to deliver applications."
As the industry shifts to a more IP-imbued world, application development efforts will drive new enhanced services and value-added applications, Nortel's Mangelsdorf says. "An open, reliable environment used to be considered mutually exclusive. We need the reliability of the telephony network, and we need open and easy-to-trade new applications."
Systems must follow the PC industry lead and be open and programmable to take advantage of complex network resources. "It can't be a couple [companies] developing applications like it used to be," Mangelsdorf adds.
The ability to support end-to-end services via IP will lead to personalized, on-demand communications. The customer must communicate with the network via point-and-click commands and enable customized voice and data applications "as personal as your fingerprint," adds Sprint's Harris. That could range from advanced caller ID and automated voice mail combinations to cameras that activate when kids come home from school and project the image--with interactive voice communication--to a parent's PC at work.
Once IP on the access side is better established, we'll see distributed subscriber management, self-service strategies and guaranteed QOS. In addition, says Terry Skemer, senior system architect at Sedona, e-commerce companies can get a boost by teaming with service providers to bolster user bandwidth while the end customer is visiting its Web site. Customers will be encouraged to return to the enhanced-services site for future transactions, he says.
Because the applications that will emerge are unknown, IP's evolution is even more critical. Open, accessible networks will spawn communications and applications to improve network usage and eliminate mass customization of applications.
The network "is a competitive tool in the marketplace," Harris says. "We move the argument from the network level to the service level. We're not talking about rocket science here, but it is a level well above where we have been in the past."
--------------------------------------------------------------------------------
Access IP
Susan Biagi
Sedona Networks has found a way to circumvent the addressing and service problems of IP. The company created the Domain Switch, which delivers IP traffic to the customer premises via a virtual wire.
President and CEO Joseph Elchakieh believes that IP must advance on the edge of the network and support end-to-end, on-demand services. "We're moving forward with a services-aware architecture to make access intelligent," he says. "Our vision is that it has to reside over the IP layer."
The Domain Switch is a hybrid of an IP router with multiprotocol label switching (MPLS). It creates multiple virtual wires--up to 2047--from the edge to the core. Each virtual wire supports a subnet with dynamic addressing and quality of service (QOS).
"A virtual wire is circuit emulation, but over IP, with well-defined jitter, delay and throughput," says Terry Skemer, senior system architect for Sedona. "The Domain Switch for the access network was created as a reaction to our struggles with MPLS. When you enter an IP network, the customer premises appears as one wire. Service can't be differentiated."
To support varying service quality in multi-tenant units, for example, the provider must run multiple IP paths to the subscriber. In addition, if a subscriber wants to connect to multiple IP networks, it must install a router to link to each one. The Domain Switch is similar to "implementing a router, without the burden of a router," Elchakieh says.
The device sits at the central office or provider's co-location site and connects to an integrated access device (IAD). The IAD supports subscriber management at the edge. Domains can be constructed dynamically, and providers can sell on-demand services based on bandwidth or QOS.
To be effective, the access network must include security and provide QOS. "IP must have the ability to support different priorities," Elchakieh says. "It is no longer acceptable to have a best-effort packet go across the network and when it gets there, it gets there. You must be able to have quality of service."
Not having end-to-end guaranteed service levels will stymie application development, Skemer adds. "If we can't guarantee it, we cannot move these fancy new services onto the access network. It won't work."
--------------------------------------------------------------------------------
A walk on the revenue side: the IP business case
Vincent Ryan
Service providers that invest now in scalable and reliable IP networks will be the winners as they launch broadband access services, according to some experts. To take part, however, executives must change the way they think about the technology underpinnings of the network, and they must reverse earlier thinking on how to build the business case.
The question to ask when considering IP investments, unlike earlier improvements in telephony, is not "Would this be a cheaper way to manufacture telephone calls?" says Larry Lang, vice president of service provider marketing at Cisco Systems. "I've seen people frame a business case using this question--they're missing the point," he says.
The correct question, Lang says, is "Given all the services I want to offer and the revenue I want to pursue, what's the correct network infrastructure?" In other words, the business case is driven not by the cost side of the equation but by the revenue side.
More than half of U.S. companies will start migrating their data, voice and video traffic to a single IP network within three years, according to a study by research consultancy The Phillips Group.
IP services will represent a $70 billion opportunity in the U.S. by 2004, compared with less than $2 billion in 1999, says Laurie Gooding, senior industry analyst of voice and data communications, of Cahners In-Stat Group (see figure).
Gooding lists e-commerce, videoconferencing, distance learning and educational networks, Web casting, multiplayer gaming, unified messaging, call centers, interactive voice response and IP-based Centrex systems as key applications from which service providers will derive revenue.
"Clearly at the forefront of the revenue opportunity is the cost savings with the wholesale IP telephony segment," Gooding says. "Large carriers are putting wholesale providers in their routing tables. Now they're turning the corner and building their own infrastructure."
But perhaps the most important trait of IP is that it's an open, market-based standard--anyone can write to it and build products around it.
"IP is going to enable services rather than boxed applications," says Jack Waters, chief technology officer for Level 3 Communications. "You used to have to be a phone company to provide apps over the network. With the platform we're building, we want to open up our network to third-party providers."
Leading-edge IP network builders such as Level 3 say revenue growth will be driven by the same demand-revenue model that rules the PC markets: silicon economics, in which halving the unit cost of computing power results in a doubling of demand.
"The early indication is that demand for telecom is even more elastic," Lang says. "Level 3 is signaling to application developers that the price of networking power is going to be on a similar sort of pace and [that they] can count on much more bandwidth at lower prices."
Although most established service providers are taking a more cautious approach than Level 3, emerging carriers are pushing the market for converged products as they look to avoid the costs of Class 5 switches and deliver quality IP services.
Lucent Technologies originally designed its PathStar access server for RBOCs that wanted to deploy second lines cheaply, but it refocused the product on the competitive local exchange carrier (CLEC) market, says James Manchester, R&D director for PathStar products at Lucent. "VCs were telling CLECs that, 'You can't come to us with a business plan that involves buying 5E switches,"' Manchester says. So Lucent built redundancy into the IP platform and added SS7 functionality and a provisioning system. "In a lot of ways, it looks like a small Class 5 switch. We built [CLECs] a converged product that they could take to Wall Street."
But again, IP economics is not just about cost. Because of the flexibility of IP technology, the delivery of IP-based services will enable all carriers to tap into larger markets and offer high-value, customized services. End users will provision services more dynamically, a feature for which enterprises are willing to pay a premium.
And the carrier's business shifts from being transport-oriented to service-oriented, Gooding says. "It's a business model centered more around serving the customer because you have the flexibility in the technology needed to do it." |
