Virtually Reliable

If there's one certainty in the uncertain world of wireless data, it's that users don't care about the hows. All they want is service that's as dependable as mom's love.

That's a tall order. The wireless environment is rough on data. Part of the problem is data's use of protocols that originally were designed for landline, where a continuous, relatively quiet connection could be assumed. Wireless can't abandon them because a significant change would force data subscribers to modify their software.

Take the transmission control protocol (TCP), which shepherds data through cyberspace by verifying that all of the packets have beenreceived and in the correct order. TCP works in the wireless environment -- just not very efficiently. It continually monitors the packets' voyage by sending additional bytes to confirm their arrival. The 40 bytes of header information per packet consumes precious network capacity.

"That's totally inefficient over wireless," said Tom D'Alleva, BellSouth director of network product management.

Packet-switched wireless also has longer transport times than landline, and those delays often cause TCP to give up waiting for an acknowledge-ment and resend all of the packets. That's a problem when the user travels into a tunnel or some other place where coverage is unavailable only briefly.

Wireless ensures reliability partly by using forward error correction, which corrects single-bit errors and detects multibit errors. The network also can use a separate, wireless-friendly protocol over the air link.

"So if there's an error that occurs there, that frame would get retransmitted quickly, and the TCP timer wouldn't have fired," said Tom Trinneer, AT&T director of wireless data applications. "You've got protection built on top of protection on top of protection. The protection fires at different levels. It's kind of like the shock absorbers in your car are going crazy, but your seat is going up and down only a little bit."

Tuning up TCP One way TCP's performance can be tuned for the wireless environment is by lengthening the period of time before it assumes packets have been lost and starts resending them, said Barry Herbert, Nortel senior manager of wireless Internet business development. That approach helps TCP cope with wireless' longer transport delays. Another approach is to use a proxy server to mediate the differences between wireless and landline.

"The proxy acts as a middleman, 'gluing' together a wireline session and a wireless session," Herbert said.

With that middleman in place, wireless can use protocols designed for its environment without forcing users to modify their applications. That's important to the future of wireless data because the more it operates like a landline Internet connection, the more receptive users will be.

Nettech's SmartIP is one such middleman that shifts part of the responsibility for reliability to the network. SmartIP also increases throughput on CDPD networks by cutting the number of packets by as much as two-thirds and allows IP-based applications to run on networks such as Ardis and Mobitex.

"The trend is that reliability is going to be addressed more and more by the network provider," said Nettech CEO Boris Fridman. "That means the integrity of the information as well as the reliability of a particular connection."

Another way to accommodate TCP is to use its cousin, user datagram protocol (UDP), which doesn't require a connection. UDP's advantage over TCP is that it only sends the packets and doesn't confirm their arrival. That approach substantially reduces packet overhead -- and earns UDP its nickname, "send and pray."

The downside is that the sender doesn't know if any or all of the packets were received unless the recipient replies. But in certain applications, UDP is acceptable. In point-of-sale, for example, the session protocol could handle reliability by monitoring the sequence of transactions and asking for any that were lost.

"You might immediately think, 'that's got to be reliable, so I'll use TCP,'" said AT&T's Trinneer. "But if the transaction is only 200 to 300 bytes, then the setup and teardown associated with the TCP session is way too much overhead. You can let the application requirements drive what increment of reliability you need."

Tweaking the Network Data ultimately is at the mercy of the air link, so how well the voice build-out addressed RF issues such as interference, multipath distortion, shadowing and poor signal coverage affects the data network's reliability.

"Even though the data users are small by comparison to the voice users, they share in the benefit of that RF engineering," Trinneer said. "That coverage is already there, as opposed to the dedicated data networks that have to grow their sites based solely on the revenues of the data business. We have a real economy of scale by leveraging it with the cellular and PCS infrastructure."

One way to improve reliability is to have a second air link as a backup. But what really attracts carriers to multiple air links are the added benefits of variable data rates and greater scaleability.

WECstar, due out by the end of 1998, chooses the air link best suited to the data. Its developer, Gooitech, is targeting automatic teller machines, which could use a high-bandwidth satellite link to download advertising tailored to each patron. Shifting high-bandwidth data to satellite takes the burden off the wireless network, which would upload low-bandwidth data such as transaction information.

"It actually works very well because those channels can also back up each other," said Chris Solomon, Gooitech CTO. "Should there be a network failure on the CDPD side, the satellite can drive the transaction."

If data has one inherent advantage, it's that like its ancestor, Morse code, it can knife through conditions that would render voice unusable. Helping data make it through is the v.42 error correction that modems use to tolerate noise and CDPD's layer-two protocol between the base station and the mobile unit.

"You might be at a level of coverage deep inside a building where you might not make a voice call, and you can still use the data service," Trinneer said. "The data network survives a little better just because it's got that robust, connection-oriented layer-two protocol."

Another way to improve reliability is to add base stations to shrink cell sites. Although that translates into higher power and improved in-building coverage, the temptation is to increase data rates.

"The higher the data rate in a radio environment, the more susceptible you are to errors and the more limited your coverage is," D'Alleva said.

Reliability also requires a data network that doesn't get sluggish when voice traffic is high. That used to be a problem with CDPD, but the situation apparently is changing. As carriers convert to digital, they can offer more reliable packet data because they have blocks of AMPS channels that are less heavily used. Dedicated data channels still can be cost-effective because it's possible to fit more than 100 data users on a single 30kHz channel.

"In a lot of cases, carriers are finding they can afford to dedicate a channel to CDPD users," said Andrew Harries, Sierra Wireless vice president of marketing. "There are also some classes of users who prefer to see dedicated channels from a performance and consistency point of view."

Central Intelligence Reliability also means continually monitoring network performance. Fortunately, many of the software tools are the same as those used by other industries.

"It's all off-the-shelf stuff, the same stuff corporate America uses for their data networks," Trinneer said. "It's not custom telco stuff. It's the data stuff that everybody buys."

Some carriers take a pro-active approach to network management. To avoid congestion and bottlenecks, they might work with data-intensive users to schedule traffic that isn't time-critical in the middle of the night.

"You really have to put some dollars into network management because your customers remember the 0.5% of the time that the network is unreliable," Solomon said. "They don't remember the 99.5%. What we need to do is get to the point where wireless networks are using network management and architecture to help each other negotiate through where the problems are."

That includes problems that aren't the fault of the carrier. If the enterprise network is linked to the carrier by an ISP, the ISP's network might be at fault. To prove that wireless data is reliable, the carrier might have to track down the source of the problem.

"We'll help them figure that out because they always assume it's the wireless link," Trinneer said. "If we don't help them out, they may continue assuming that."