ISP Technologies: Choosing Which Knives To Juggle
By Peter Lambert
According to the experts, Internet service providers will face critical technology decisions over the next year, but tough strategic calls must be tackled first. Just which slice, or slices, of the networking business will one pursue?
Specialization increasingly rules the day, and some providers now object to even being called Internet service providers, a term that connotes a generic Internet access play - a substantially different goal from specializing in distributed World Wide Web hosting services for businesses, for example, or from bundling voice, video and data services to consumers and businesses alike.
Savvis Communications Corp., for example, describes itself as an "Internet carrier" with a unique, high-performance backbone infrastructure. Digex Inc. uses state-of-the-art database and server facilities to provide "complex Web hosting" services. Exodus Communications Inc. and GlobalCenter Inc. market content distribution and emphasize their geographically distributed data centers.
Yet some common technical challenges persist for both specialist and generalist:
How does one stay affordably ahead of demands for network capacity?
How does one best guarantee consistent performance for customers?
What technologies will enable the smoothest launch of value-added services as demand requires?
How many facilities does one build and where?
Service providers "need universal access for customers, more bandwidth, service classification and security, and the ability to support virtual private networks, IP fax and 20 other high-value applications coming over the next several years," says Steve Thomas, director of Internet marketing at Ascend Communications Inc.
And they need to provide quality, according to Dan Minoli, director of engineering and development for Teleport Communications Group (TCG), a competitive telephone and data services provider.
"What we have to worry about is the ability to sustain performance through each new level of demand," he says. "That will require guaranteed quality of service for data, voice, video and multimedia; flexible billing systems; and access technologies that must grow from thousands to millions of lines."
Continued stratification of ISP business models will affect their technology needs. The new common wisdom: Providers must offer unique and better services to escape cutthroat, low-margin, commodity price wars.
Focus First
"I don't believe any ISP will be living on $20 per subscriber this time next year," says Ascend's Thomas. He projects a services food chain that includes a handful of tier-1 backbone providers and thousands of local access providers, plus middlemen in the form of "wholesale [wide-area router and switch] port sellers" and "virtual ISPs" that integrate end-to-end services running over other companies' facilities.
Such specialization requires special technologies. For example, a specialist in virtual private network services will need to deploy layers of security technologies, including encryption, firewalls, Remote Authentication Dial-In User Services servers and remote-user smart cards with constantly changing passwords - all to guarantee enforceable access to private information over public networks.
A specialist in high-performance backbone access needs to deploy routing, switching and circuit load-balancing technologies for the best connections to the greatest number of backbone providers.
A specialist in complex Web hosting and content distribution must deploy the latest content replication, transaction processing and other database and server technologies.
Such specialists become the target market for a new generation of managed services, says Glenn Falcao, vice president and general manager of Northern Telecom Inc. Public Carrier Networks.
"The intelligence sits in the public network, at the edge of the network, and is managed by the service provider," he says.
The ISP buys a service and the carrier manages the modem bank and the required network interfaces. A new generation of access equipment from multiple vendors will support that capability.
Supporting unpredictable access growth will be no mean feat in 1998. According to vendors, the primary trick will be rapid deployment of interfaces of varied types and speeds, requiring that service providers deploy a universal modem bank that can be easily expanded line card by line card.
The hardware in those universal access concentrators also is increasingly modular to assure they don't hit a performance ceiling as more and more line cards are added. The primary test lies with whether the vendor's design uses dedicated processors in each line card. Most suppliers have achieved, or are developing, this architecture, so that no central processor bears the burden of every new access port added to the rack.
For managed business network services, a new breed of multiservice access switch has emerged from frame relay and ATM switch providers, including 3Com Corp., Ascend, Bay Networks Inc., Cisco Systems Inc., Fore Systems Inc., General DataComm Inc., IBM Corp., Lucent Technologies Inc., Newbridge Networks Corp. and Nortel. Those switches are designed to translate any customer's mix of frame relay, X25, Ethernet or ATM campus protocols into wide area data links.
The service provider that can turn up a new customer service line, accommodating any in-house enterprise protocols with the addition of a simple switch component, will compete well.
Capacity Challenge
As of late 1997, major backbone providers say their networks now offer capacity 14 times greater than a year ago, and, according to some, this year brought a sea change from reactive to proactive capacity planning. Backbone providers now plan two to three years out, rather than six months.
Such a sea change must be reflected at the edges of the Internet, too, and, indeed, backbone providers are requiring higher capacity and more points of presence of those that connect to them.
Service providers such as Exodus Communications Inc. and InterNex Information Services Inc. no longer simply react to increased demand by adding more 1.5-megabit-per-second or 45-Mbps access lines to their backbone connections.
Instead, they are deploying new classes of very high-speed switches and routers, such as Internet Protocol (IP) Switches from Ipsilon Networks Inc. and Route Switching Modules in their Cisco Catalyst ATM switches. Both technologies move route processing out of slow software into fast silicon chips.
But, says Jonathan Heileger, chief technology officer for digital distribution provider GlobalCenter, "the technology to support large pipes is still in an early phase of development. It's normal to be on the bleeding edge of development, but it's often not enough. The growth struggle has been negotiating with all of the ISPs who are adding consumers nonstop."
As those consumers request more traffic from content providers hosted by GlobalCenter, "we mutually need to ensure that we have great connectivity between us - whether it's a little ISP in Georgia or a major carrier like Sprint [Communications Co.] or MCI [Communications Corp.]," Heileger says.
To that end, he says, GlobalCenter and others have "developed a lot of in-house tools to aid us in traffic management, load distribution and server management." Such cooperation and demands for software feature development require more and more rarefied expertise. Technical hires will increasingly matter more than supplier decisions.
Smart Bandwidth
According to TCG's Minoli, just as a new jet engine technology, beyond bigger and bigger propellers, was needed in 1944 to make airplanes more powerful, so a technology leap is needed to make networks not only faster but more intelligent.
Call it quality of service (QOS) or class of service. Either way, providers increasingly argue that survival will depend upon the ability to shape bandwidth, distinguish between high- and low-priority traffic and keep promises to customers concerning network performance specific to each class of traffic.
TCG's fundamental QOS technology decision: Route at the edges, but switch in the wide area to replace multiple, performance-stunting router "hops" with single-hop, dedicated circuits.
"We believe we must not be forced to unwrap and rewrap packets through the network, so we use ATM in the core, the only switching technology available with quality of service mechanisms," Minoli says. "ATM is necessary beyond 12 nodes, and, over 500 miles, ATM economics win."
Like TCG, Savvis and other national providers lease their ATM circuits to local ISPs, as well as businesses, to ensure dedicated long-haul capacity with minimal hops to maximize performance.
However, other providers may remain content to stick with Ethernet-only networks, thanks to technologies designed to make Ethernet more bandwidth-intelligent. Ethernet router providers continue to develop both faster routing hardware and "cut-through" flow control software to maintain forwarding speeds equal to switched circuits.
Further, Ascend, Bay, Cisco, Motorola Inc., Paradyne Corp., Sync Research and other router and frame relay access device makers are implementing ATM-like features that can prioritize each data session according to its need for bandwidth. In times of router congestion, the router "reads" the high- or low-priority tags on a packet to determine the packet's place in line at busy port.
This session prioritization can approximate dedicated circuit QOS, translating to a service provider's ability to guarantee throughput for some services, even in the "best-effort," send-and-pray environment of Transmission Control Protocol/IP over Ethernet.
A number of third-party software providers also have begun to commercialize circuit and server load-balancing tools. Suppliers including F5 Labs Inc., Packeteer Inc. and Xedia Corp. are delivering devices that prioritize traffic into classes of services based on application type, enterprise department or individual user.
In terms of distributed design, providers such as Bright Tiger Technologies Inc. and Inktomi Corp., Network Appliance Inc., Silicon Graphics Inc. and Sun Microsystems Inc. are delivering tools that automate the process of replicating, updating and synchronizing a Web site across geographically dispersed Web site location centers - products driven by the new wisdom that the bandwidth intelligence quotient goes as much to distributed network design as to priority or QOS tools.
Such distribution puts content closer to users, thereby raising performance of the content site. Simultaneously, such designs take pressure off long-haul backbone bandwidth and provide automatic fail-over from a broken or underperforming server to an available server with the same content.
The efficiencies of a distributed network architecture also go to connections with other providers.
Savvis Communications says it came out on top of this month's Keynote/Boardwatch Internet Backbone Performance Index measurements because of its one-to-one private Network Access Point (NAP) architecture, which bypasses all crowded public NAPs to deliver direct connections from Savvis to each backbone provider.
Similarly, Exodus - which early next year will add data centers in Boston and London to its data centers in Los Angeles; New York; Santa Clara, Calif.; Seattle; and Washington, D.C. - is building each data center close to Internet exchange points to minimize traffic hops and optimize performance.
"Adding bandwidth is not always the right answer," says Alan Leinwand, chief technology officer for Digital Island Inc., a Honolulu-based provider that has created direct, single-hop connections to backbones in 25 nations. "Simplified topology can translate to quality of service."
Fleet Of Foot
Going forward, providers may have less lead time to anticipate customer needs.
According to David House, chairman, president and chief executive officer for Bay Networks, enterprise networks now define the real organizational structure - "how people actually work" - more than do departmental definitions. If, for example, a product design department project requires that the company launch Extranet services to its business partners, the ISP must quickly get its hands on conditional access technologies that allow only specific users into specific segments of the customer's private network.
And, according to Jeff Ralston, vice president of engineering for Four11 Corp., an Internet white pages service acquired last summer by Yahoo! Inc., product differentiation will beget more and more technology knives to juggle.
Four11 wants to see network-based backup, traffic load-balancing, redundancy, service classification, rapid capacity deployment and multiple security tools from its providers. Most of all, Ralston says, the company wants "smooth integration of new [Four11 service] features with the access provider's network."
Soon, the same may be true for millions of small residential businesses and even consumers getting onto the Internet. If one ISP fails to respond to sudden new demands, another may be ready to fill the bill.
|
