americasnetwork.com
urry up and wait." Not only is that expression the credo of Internet users, but it's the mantra of Internet service providers (ISPs) as well. While local exchange carriers (LECs) and cable TV (CATV) operators tentatively deploy asymmetrical digital subscriber line (ADSL) and cable modem services, respectively, ISPs are left trying to satisfy users' insatiable appetite for bandwidth.
Apart from 56 kbps modems, the only other high-speed alternative ISPs have to offer is integrated services digital networking (ISDN), a service that has acquired a (some say undeserved) bad reputation in the United States. So, since they don't control the facilities, ISPs and their subscribers have had to sit and wait for the newer, faster access technologies to come to them.
However, ISPs might now have a chance to do something for themselves. Channel bonding, an established bandwidth-on-demand technology for ISDN, has been tweaked to bond twin 56 kbps analog modems for a potential 112 kbps throughput; something that could put a smile on the faces of ISPs and users.
Bonding's beginnings
Channel bonding isn't gleaming, new technology. In fact, it's been an ISDN offering for quite some time (hence the name "channel").
In channel bonding for an ISDN basic rate interface (BRI; 64 kbps) connection, when a call is initiated, it brings up the first B channel. Then based on a pre-definied bandwidth threshold (set by either the user or the ISP), a second B channel will be brought up. So users have either 64 kbps or double that, depending on what they need.
"You pop up to that second connection or you pop back down if you're not using it," explains Kurt Bauer, vice president of access product management for Ascend Communications Inc. (Alameda, Calif.). "That value of it in the ISDN world has been that [the service provider] got paid for usage of that second B channel, so [users] don't want it up all the time if [they] don't need it."
ISDN channel bonding can run all the time; however, that costs ISPs and users more money, Bauer says.
Usually, channel bonding protocols are built on top of point-to-point protocol (PPP), using multilink protocol (MP) or an extension called "multilink protocol plus" (MP+), which was developed by Ascend, but is an open specification, according to Bauer.
"The way PPP gets established, there is a single IP address per connection," Bauer continues. "If you want to bring up two lines using MP, then PPP says that you have to tear down the whole link to bring both of them up. That's not very nimble, and it's a lot of overhead.
"What MP+ does is allow that second channel to be toggled without deconstructing the link on the first channel," Bauer says. "Before MP+, that was not possible. In order to turn up that second line, you'd have to drop that first line and then redial and bring them both up. It was really clunky."
Also, without a smooth addition of a second line, real-time applications go out the window. "The whole user experience is `hurry up and wait,' " he says. "MP+ fixed that."
Now, just change the modems
Yet, ISDN isn't always easy for ISPs to offer, or for users to get. Not only can line qualification and service provisioning turn into an ordeal, but the prices aren't exactly what many home and small office users would call economical. Also, there can be a fear factor for non-technical users when installing an ISDN terminal adapter.
"Many parts of the country and the world can't get ISDN service, or there are people who just don't want to deal with ISDN-it sounds complicated even though it's not, really," Bauer says.
So, MP+ was adapted to perform the same kind of bonding, but with analog modem lines, rather than B channels (Figure 1). In a typical application, a user would have two analog lines connected to a dual-line modem (ideally, a 56 kbps device), which is connected to a PC via a serial line. Software on the PC is used to configure the modem with the traffic thresholds that signal channel bonding to begin. A user connects to the ISP; as the bandwidth he or she requires meets those thresholds, the second channel goes up.
Figure 1: Ascend's analog modem channel bonding vs. ISDN channel bonding
"Based on the traffic volume of the first link . you bring up the second link with the same IP address," Bauer says. "So you've got an aggregate throughput of 112 kbps, theoretically. It's fundamentally the same as what happens on the ISDN side. It's not as fast as ISDN will be, but it will definitely work."
Also, analog channel bonding is speed-independent. It can work just as well with a v.34 modem as it can with a 56 kbps modem.
"It's just more interesting when it's a 56K modem," Bauer says. "Then when we all upgrade to v.pcm [the International telecommunications union's new 56 kbps modem specification], we'll all be running with a unified spec."
Can it deliver?
Obviously, 112 kbps of throughput via two bonded 56 kbps modem lines is a "best of all cases" scenario. Law limits analog modems to 53 kbps, and outside noise, line taps and all the other hazards of copper plant ensure that results will greatly vary. The outcome is that running the copper gauntlet means that 56 kbps modems can't offer 56 kbps. Even at their best, they may be signficantly slower. The same holds true if they're bonded.
"I think the best people have seen is around 48 kbps outside of laboratories," says Patrick Kloepfer, director of remote access services for GTE Internetworking (Cambridge, Mass.). "Typically you see around 40K to 42K." Beyond copper limitations, 56 kbps modems aren't symmetrical. At best, they offer 56 kbps downstream and 33.6 kbps upstream, he adds.
Still, channel bonding for analog modems could provide benefits, Kloepfer says, especially if ISPs use it with 28.8 kbps or 33.6 kbps modems.
"What would be interesting is if people said, `Hey, instead of doing 56K, which isn't giving us any better throughput, maybe we should just bond two 28.8s, which give good reliable throughput, and get a true 56,' " Kloepfer says. "The bonding of two 56Ks almost seems strange. If you're trying to get better throughput, you're almost better off justifying an ISDN circuit."
Also, there are issues for ISDN channel bonding that Kloepfer says would most likely apply to bonding analog modem lines.
For ISDN channel bonding to work, the caller, when establishing the new channel, must land on the same device for them to be bonded. In less populated areas, this may not be a problem, since statistically it's highly probable that a user's two calls will hit the same device. That doesn't ring true for urban areas.
"If you're supporting New York, where you have 10,000 modems, that's 100 boxes of 100 modems, the statistical probability of you landing on the same box is 1%," Kloepfer explains. "So, effectively, you don't have a service offering bonded ISDN. I'm going to suspect that, for any new bonding system to work, it's going to have the same limitations. It must be on the same box for it to bond."
Even if the statistcal probability of hitting the same box is one in three, two-thirds of the time users will be calling in to complain, Kloepfer says.
With regard to the complexities of ISDN vs. analog modem channel bonding, Kloepfer questions whether, when users have to buy two analog lines (chances are, if users are savvy enough to want this speed, they'll be on their third) and configure their modems with specific traffic thresholds, there's no real difference between doing that and subscribing to ISDN services.
"I think the answer you'll find is that ISDN can be more expensive," he says. "But maybe the solution is not to come up with alternative ways to busy out the nation's phone system even further. You'd be better off going to the phone companies and saying `make ISDN affordable.' "
Don't write it off just yet
Still, some ISPs see a reasonable business model for offering such a service. EarthLink Network (Pasadena, Calif.) is testing (on a board level) Diamond Multimedia System Inc.'s (Vancouver, Wash.) Shotgun technology, which is based on Ascend's MP+, according to Barry Friedman, EarthLink's director of product management.
So far, the tests have yielded a dual-analog connection at 112 kbps and transfer rates constantly exceeding 80 kbps, Friedman says. The tests were conducted over a connection on public switched telephone network (PSTN) lines, which was dialed in long distance from Vancouver, Wash., to an access number in Southern California. EarthLink was set to begin internal testing by press time, and plans to conduct remote testing with various modem makers.
EarthLink doesn't have to do much in the way of infrastructure upgrades to offer services. It just has to shake out the bugs, Friedman says.
"We use Ascend MAX and TNT boxes, and basically all that changes on our end is a revision of the Ascend software that allows this to occur," he says. "There's no hardware changes on our part. There's a software upgrade, and obviously there's all sorts of production issues before you release new software under your production POPs [points of presence]."
Also, EarthLink will have to ensure that its backbone providers (EarthLink doesn't own its own backbone outside of California) can support the technology. At least one of the providers will have to join the testing, Friedman notes.
Apart from the rigors of testing, Friedman doesn't see nearly as bleak a deployment scenario as Kloepfer expects. As far as the demands channel bonding places on POPs, the impact of having users taking up additional ports isn't easy to pinpoint, he says. Obviously, when users are channel bonding, they are using two ports, but they will most likely have far shorter Internet sessions.
"[Channel bonding] takes two ports, absolutely," Friedman says. "When you run out of ports, you run out of ports. You don't achieve the second connection. The bigger issue, is if I have 10 ports and eight people connecting at 28.8 to download 4-megabyte files, would they not be better served if they connected dual-channel at 80 kbps, downloaded those files and logged off, freeing up those ports quicker?"
Also, a bonded connection isn't constant; it becomes active once a threshold is met. Based on EarthLink's experiences with ISDN users (who usually subscribe to a certain number of channel hours) subscribers generally stay under their time allotment for bonding to avoid paying additional fees, Friedman says.
Earthlink will price its dual 56 kbps service between an unlimited access dial-up account ($19.95) and an ISDN account ($34.95 for 100 channel hours with 99 cents for each additional), according to Friedman.
"I'm looking at something around $29 a month with 100 channel hours," he says. "You're paying a $10 premium for the ability to bond two channels. If users find that they need the bandwidth and the time, we'll charge them a surcharge to do that, because at that point it really does start to impact my ability to manage my POPs."
Shelf-life
Assuming that channel bonding is a viable option, ISPs must also consider its staying power. Can bonded analog modem lines compete with the likes of ADSL and cable modems as those technologies become real services?
"We already sell cable product, and that is a very viable business and consumer offering where it's available," Friedman says. "We're doing DSL trials right now. I see dual 56K as a viable technology, especially for small businesses that already have two phone lines installed. The modems that are shipping have the same logic as terminal adapters, so they have bandwidth-on-demand capabilities. They won't tie up both lines and result in a busy signal. People won't have to order an ISDN line and all the confusion people think is associated with that."
Also, if copper lines are bad, such as in a remote rural area, and 56 kbps service is but a dream, Friedman sees Kloepfer's scenario of bonding 36.6 kbps or 28.8 kbps modems as a viable service offering and a worthy competitor to ISDN BRI service, which may not even be able to reach that customer.
Friedman is optimistic. "There's more of a market for dual 56K than there ever was on ISDN or there ever will be-on the short run-for DSL or cable modems |
