Though Web services have been much ballyhooed for their ability to reduce the amount of manual coding for programmers, glaring inefficiencies in the way XML (define) is digested threatens to paralyze distributed computing.

Modern-day Web services are largely based on XML, which analysts say is fine for simple functions, but tends to get bloated when used in massive quantities.

Current XML parsers are not very effective. The APIs (define) have just as much trouble reading large file types in an efficient manner as they do reading multiple small file types.

This stresses out current processors, which power the application server to deliver content. Research firm ZapThink said this poses a problem because it believes corporations will continue to ramp up the amount of XML they employ in their networks, expanding from 15 percent today to almost 50 percent by 2008.

Because it is believed that Web services traffic will dominate XML traffic by the end of 2005, the possibility of a network bogging down at crucial points during, for example, a purchase order execution, increase 10-fold. This is a horrifying proposition for the business that relies on the Internet to fuel its money-making transactions.

With the glamour of Web services steeped in the possibility of processing thousands or even millions of transactions on a network, the threat that insufficient XML consumption could tie up computer systems is very real. ZapThink analyst Ronald Schmelzer said customers and vendors have expressed concern about XML's ability to underpin Web services.

He admitted the problems with XML processing put a damper on the research company's prognostication for the multi-billion-dollar growth for service-oriented architectures and distributed computing.

After all, ZapThink believes the market for SOA (define) systems will balloon from $4.4 billion in 2005 to $43 billion by 2010.

But, he noted, several vendors are rising to meet the challenge. One might think that Microsoft, IBM , BEA Systems and others would lead the innovation of the so-called XML optimization market. Not so.

Optimizing XML

Companies like DataPower, Tarari and Xambala are providing chipsets or cards that can be used to clear the path for speedy XML digestion on networks. Others, like Rogue Wave, are focusing on fixes in the software stack to deliver them from the glut of fat XML files.

Forrester Research analyst Randy Heffner believes that, while the inefficiencies of XML impact a number of network functions, the problem is at the processor level. Even adding a basic function like encryption can knot up Web services, he said.

That's where companies like DataPower come in. Eugene Kuznetsov, co-founder and CTO, was one of the first major engineers to relay the importance of XML optimization to the industry.

Since its launch in 1999, DataPower has created products for speeding up, securing and integrating XML. Initially, Kuznetsov said, the company tried to treat the problem with software.

But it quickly realized that having its own chipset was the key, because current processors aren't equipped to handle massive loads of XML. That's why the company created the XG4 chipset to process XML faster and sold it in an appliance along with the software.

"There is a big tradeoff with XML," Kuznetsov said. "You get all of this functionality, ease of implementation and interoperability at the cost of performance. Large financial services institutions have this problem in spades."

In contrast, Rogue Wave Software, a division of Quovadx, provides advanced processing via software, according to President Cory Isaacson. His company subscribes to the notion that native code is better for processing XML and Web services than languages like Java or .NET.

Rogue Wave has created a product called Lightweight Enterprise Integration Framework (LEIF), a language-independent tool that does the heavy lifting of XML message management using native code, or code that is compiled to run with a particular processor. LEIF takes C++ applications and exposes and consumes them as Web services in native code.

Isaacson said Rogue Wave is planning similar solutions for languages other than C++ but declined to elaborate. While Isaacson acknowledged DataPower as an early mover and that hardware solutions will have a place in the networks, he said native code has an advantage.

"Even if you parse a message faster in a hardware set, it doesn't necessarily give you any access or intelligence for what to do with the message," Isaacson said. You still have to have application code that understands what you want to do with the message to really make it efficient. We think that native code is going to have to play a part in this."

Heavyweights Slow to XML Optimization

ZapThink's Schmelzer said the glut of XML is thick enough to bear a large market in which the likes of DataPower and Rogue Wave can sell their wares. He predicted the XML performance optimization market will reach $1.2 billion by 2010.

That could be a conservative estimate. After all, Web services purveyors like Microsoft, IBM and BEA haven't made any large splashes into the pond -- yet.

Forrester's Heffner said optimizing XML is on these companies' radar for sure.

"They see the need and if it fits in priority scheme, long term they will have this integration," Heffner said. The question is how they go about it: will they put it in front of the server or in the server?"

Last month IBM, Microsoft and BEA proposed the XML-binary Optimized Packaging specification to the World Wide Web Consortium following a workshop on the topic in September.

So, although they don't have products, they are certainly weighing the dilemma that faces the industry. How can these vendors sell customers Web services infrastructures if they can't provide them with fast, multiple messaging functionality?

And right now, simple Web services have been enough, even though Heffner and Schmelzer allow that add-ons like the Web Services Security stack will choke the network some more.

"There are enough scenarios where the latency is not that critical that it could take some slower processing of XML messages," Heffner explained. "It's when the volume increases and things start backing up that it becomes a problem."