Souped-Up

Wireless data often gets a bum rap for being too slow. But middleware tweaks can make 14.4kb/s wireless feel just as zippy as 56kb/s wired.

P. K. Prasanna has a laptop, a CDPD modem and a few gripes about his experiences with wireless data.

"It's incredibly slow; it's incredibly volatile," he said. "On some days, it takes maybe 90 seconds to download a Bloomberg page, and other days, you can't get to it no matter what. The main problem with CDPD is that it's very erratic. The delay is unacceptable. Even though the throughput is advertised as 19.2kb/s, we've typically never achieved throughput of more than 8.5kb/s."

Ouch. If there's an upside to that assessment, it's that Prasanna isn't a typical data user. He's also president of Flash Networks, one of several vendors trying to improve throughput, a key performance factor that determines user experiences, which in turn determine market acceptance.

Assessing throughput and latency often involves building automated applications that periodically track and report network performance.

"We've seen service providers increasingly using user agents: An automated app that uses a user agent, which is transitory, that runs through typical end-user scenarios, measures response time and then feeds those response times into the application that's collecting this info," Prasanna said. "For example, to measure the performance in a dial-up wireless ISP scenario, a Web browser would be scripted to dial up the network periodically, perform a set of typical page, file and e-mail downloads, and record the results. In some cases, we've seen carriers also install transient monitoring agents on client devices that are there for the duration of the session and then self-destruct and report back."

Too Much Overhead
One reason for less-than-wired throughput is that TCP is inefficient in a wireless environment, thanks to its constant get-it?-got-it acknowledgement that accompanies packets to ensure that lost packets are retransmitted. Middleware cuts that overhead through approaches such as substituting much more efficient protocols, such as UDP, which doesn't send acknowledgments.

"We've built a protocol that's TCP/IP-based, but it uses UDP to transport information," said Patrick Glenn, Fourelle chairman & CEO. "We built a TCP-like reliability on top of that, but we use very efficient means."

Each vendor's approach varies, but in a typical deployment, the middleware resides at both the user's device, such as a laptop, and a server, which could be in the provider's network or in a corporate network. When the packets from the user's application reach the middleware, the TCP is translated into an efficient, often proprietary protocol before beginning its journey over the air link. At the other end, the process is reversed.

Although the goal is to reduce each data transmission as much as possible so that the "pipe" can carry as much as possible, each vendor takes a slightly different approach. Broadbeam's Smart IP, for example, sends the acknowledgment only at the end of the message.

"If some packets are missing from the message, instead of sending the entire message, you can resend just the missing data," said Tamara Gruber, Broadbeam director of marketing. "It depends on the type of application, but in general, you can look at improvements of 50%."

Just the Facts
Another way to slim data is through compression, which can give users the impression that the service offers 56kb/s even though the air link actually is maxed out at only 14kb/s. In a sense, compression is like WAP: It massages the Internet so that it's more in tune with wireless' limited form factors and bandwidth.

"What we end up delivering to the user is, on average, between two and three times the speed associated with the 14.4kb/s connection," said Brad Fisher of Bell Mobility, which offers circuit-switched CDMA. "That's strictly for Web content, POP3 e-mail and FTP."

Compression's effectiveness varies by the type of wireless-data application.

"Some graphic images on the Web are more easily compressed than others," Fisher said. "Some pages will come down five times as fast as they would have without compression. Some come down two times as fast. With our extensive testing, we've seen an average performance enhancement of 200% to 300%. It really makes the wireless Web usable. We got comments back (such as) 'Feels like a landline connection' and 'Couldn't tell the difference.'"

Indeed, latency and throughput are factors that are easily perceptible to the user when accessing a Web site or corporate intranet.

"That's what keeps the user glued to the screen or lets the user walk away because it's taking too long," Prasanna said.

Other performance-related factors include:

• RF engineering. Although data tends to be more robust than voice, spotty coverage, weak signals and interference still can have a significant effect on performance.

"As the S/N degrades, the number of errors increases, the number of retransmissions will increase, which will decrease the throughput and increase latency," said Peter Rysavy, Rysavy Research president.

• Forecasting. Regardless of whether the rosy predictions of wireless data's growth come true, understanding how different applications and more users affect latency and throughput helps ensure that supply can accommodate demand.

"Because we sell applications and not just the 'pipe,' we have a good idea of how much bandwidth (will be needed) and what the call models are for our applications," said Tom Bedics, Southern LINC manager of technology development. "So it makes it a little easier for us to model what we should expect as we add customers."

That raises another point: Like TCP, many applications designed for use in wireline networks don't always adapt well to the wireless environment. The bottom line is that the user experience, good or bad, stems from a lot more than just the maximum raw throughput that the network can offer.

"It has a lot less to do with what the carrier does than with how the application is designed and deployed," Rysavy said. "A lot of apps simply won't func-tion at the data rates provided by wireless. That's why things such as WAP were invented: A lot of things just fall on their face when you try to run them over the pokey speeds we have today."

---

Don't Fly Blind

Many wireless-data services provide access to servers in another network, such as a content provider or a corporate intranet. That approach means additional factors that potentially affect throughput and latency. For example, if subscribers complain about slow response times accessing a particular site, ferreting out the problem means determining whether cell sites are at capacity or whether a server on a partner's network is bogged down under too much traffic.

Although service-level agreements are common in telecom, they're still a rarity in wireless data, partly because there are so many wild cards, such as spotty coverage and interference, that can affect performance. Even so, a combination of technology and a close relationship with the other network can help ensure that potential problems don't go unnoticed.

"We have network-performance monitoring that tracks various links," said Chris Langdon, Telus Mobility director of marketing, wireless-business unit. "We can measure all the way up to the customer's server. With the Phone.com UP server, there are error messages generated if the terminating server is unable to respond to the request. Once we see that, we get alarmed. Depending on our relationship with the service provider, if we have an exclusive arrangement and negotiated performance standards, we'll generate a call from our network folks to their people."

But that situation could get difficult as wireless offers access to more sites and, eventually, the entire Internet.

"With microbrowsers, people are going to other Web sites that we don't have any relationships with," Langdon said. "In those cases, it would be like going to a site on the wired side, where the server is down; there's generally no troubleshooting follow-up for the ISP."