Speed Has Its Price

by Jim Thompson 

July 14, 2000 --  Shakespeare warns of the man who is "hoist on his own petard." A petard was an early explosive device, designed to be set against gates or other barricades for the purpose of blowing them to bits. 

As often happens in the early history of a technology, there were glitches. Unlike today, when the worst thing that can happen is the CEO of a new company gets stuck with a lot of worthless stock after a wild IPO, the petard carried a far more lethal fate. In this case, the poor sap delivering the explosives was himself often blown up. There's a certain irony, no doubt lost on the victim, in being blown up by ones own weapon; hence the phrase. 

Don't blame the user
For years the 'swimming-in-molasses' sensation that was a part of most people's Internet experience has been blamed on end-users. Their 'old computer and dial-up connection slowed everything down' was the popular notion. 

But with the current marketing explosion of high-speed access brought on largely by the proliferation of DSL, ISPs are suddenly finding that they can no longer blame the end-user for the slowness of the Net. The tables have turned and now, the end-user is blaming the ISP. 'I have a high-speed connection, so why aren't things faster?' is the question. 

The irony here is that it's been the ISPs who brought about the revolution in high-speed access by marketing DSL. I haven't seen many petards around, but, in the distance, one can hear a distinct 'Ka-boom!' 

This slowdown of the Internet is highlighted in a recent report by the investment banking firm of Bear, Stearns & Co. Entitled "Internet Infrastructure & Services," the report highlights several places on the Internet where information flow is slowing to a snail's pace. It asserts that the growth in end-user, high-speed access is only accentuating the inherent slowness of the Net. Of the forces bringing this weakness to light, probably the most powerful is the expansion of DSL deployment. 

Explosive DSL growth
DSL lines in service in North America passed the 880,000 mark at the end of the first quarter 2000, according to statistics released by TeleChoice, Inc. The number represents an impressive 50 percent overall increase in U.S. deployed lines over the end of 1999. 

The continued deployment of DSL is an important factor in the growth of the Internet. It's also a significant part of the revenue stream for ISPs and CLECs. Even ILECs are starting to get into the game. (ILECs currently lag well behind the CLECs in high-speed DSL deployment, but they're rapidly making up for lost time. First quarter 2000 results show them aggressively addressing this issue.) 

"It was the challengers like Covad, Northpoint, and Rhythms that were exceeding expectations, while the ILECs were coming in at, or below, expectations. But the ILECs are now beginning to change that," said Bear, Stearns Internet infrastructure analyst, and the author of the study, Robert Fagin. 

Easier DSL deployment
A major push is underway to take the difficulty and the cost out of deploying DSL lines, which is also adding to the growth of the technology. In early July, SBC Corp. announced a do-it-yourself DSL modem package that they claim the customer can hook-up and put on-line themselves in about an hour. 

With a downstream speed of 1.5-Mbps, it sounds a lot like G.lite. However, to date, SBC hasn't made any announcements about the type of equipment they will use for what they call 'Basic Internet Service.' Northpoint Communications is hoping to meet the challenge with its rollout of G.lite. Northpoint began testing a G.lite modem last year and has been test-marketing G.lite modems through Radio Shack stores in the Dallas/Fort Worth and San Francisco Bay areas. 

Despite this proliferation of theoretically high-speed access, however, things are getting slower, not faster, for the average user. So, what's making it so slow? 

Internet choke points
In what author Robert Fagin calls a "holistic approach" to the problems of Internet data transfer, the Bear, Sterns report identifies primary trouble spots: congestion at the network level, whether that network is public or private, congestion at the backbone level, and the previously mentioned, and much abused, end-user 

The proposed solutions include adding "intelligence" at the network level. At the backbone level, the report suggests serving content from locations as close as possible to the requester and optimizing routing. 

Content owners are finding they can't solve their problem simply by adding more servers and/or bandwidth. Adding more servers to deal with peak demands results in those servers sitting idle during periods of low demand. Bandwidth is like real estate in Hawaii, there's never enough and the price is always rising. 

To keep up with demand, content providers are turning to intelligent solutions like caching, load balancing, and bandwidth management to relieve network congestion. 

Caching technology stores the most frequently requested portions of a Web site on a separate server that is as close as possible to the requesting user. Local Internet storage caching is less expensive than network retransmission by about 40 percent per year. While not all Internet traffic is cachable, as much as 50 percent of the majority of all Internet content is cachable. 

Since many Web sites these days are made up of clusters of servers, load balancing, which works by routing requests to the server within that cluster with the most available capacity, is another way of helping to speed things up a bit. 

Congestion occurring at the backbone level is due to an inherent flaw in the architecture. "Since most Internet data is exchanged between networks for free, backbone providers have an incentive to invest in intra-network performance, but not in inter-network efficiency," noted the report. 

Backbone congestion
Because the money is in making one's own network run better and not in Internet efficiency as a whole, backbone providers want to get rid of data packets not bound for their own customers as soon as possible. That means they dump packets at the nearest point without regard for eventual destination or network congestion. 

Solutions to the problem of backbone bottlenecks are concentrated around trying to avoid the public Internet network as much as possible. One solution is offered by the Content Distribution and Streaming Media Networks (CDNs/SMNs). Companies like Cidera, iBeam, and Akamai are examples of companies using this approach to making things faster. 

InterNAP approaches the network congestion problem through an optimized routing solution. "They do all the interconnects by themselves, which is part of the beauty," said Fagin. 

"They buy high speed access from UUNet, AT&T, or Cable & Wireless, and then they take care of the connections internally. They've got these big banks of routers which basically say, 'the user is coming in on the Qwest backbone and the request is to somebody on the UUNet backbone. Let's re-route the request ourselves for a fee, rather than leave the Internet to its own devices.'" 

ISPs to the rescue?
But all of these attempts at speeding up the Internet are little more than patches. The Net remains slow. The rapid implementation of DSL along with other high-speed technologies has only exposed the weaknesses in the Web. 

Unless they can find a way to make it profitable, backbone services don't appear willing to fix the problem. Solutions offered by companies like InterNAP or the CDMs/SMNs are oriented around getting away from the public Internet, so they offer little essential help. 

It comes down to ISPs and content providers to do something about the slowness of the Web. Why? Because the success of their business is based on the overall quality of the Internet. Quality means increased usage, which translates directly into increased revenues through value-added services. 

ISPs are standing by the gate, petard in hand, hoping to bring down the walls and live to tell about it.