Breaking XML to Optimize Performance

As XML becomes ubiquitous throughout the enterprise, it increasingly taxes the systems that must deal with it. Even though there are a wide range of hardware and software solutions coming to market that aim to alleviate XML’s performance bottlenecks (See ZapThink’s XML Proxies Report), many developers are nevertheless resorting to a variety of tactics to improve the performance of XML processing and transmission that are… well… creative. Many of these creative approaches simplify certain aspects of XML in order to squeeze document size, improve parser performance, and speed the mapping of XML document components to application objects.

Why must developers jump through so many hoops to improve the performance of XML? Simply put, XML is not a particularly efficient format for representing information. It is a text-based, human-readable, and metadata-encoded markup language that operates on the principle that the metadata that describes a message’s meaning and context accompanies the content of the message. As a result, XML document sizes can easily be ten to twenty times larger than an equivalent binary representation of the same information. Even though it is inefficient, however, XML’s numerous advantages are increasing its use for ever broader and more mission-critical functions. (See ZapThink’s popular Pros and Cons of XML report for a detailed discussion of the merits and challenges of XML).

While XML’s verbosity may be acceptable for situations with moderate transaction volumes, XML’s processing overhead, storage requirements, and bandwidth consumption become quite problematic when transaction volumes are high. As a result, many companies are resorting to potentially dangerous tactics for squeezing every last drop of performance out of XML. Three common tactics include compressing XML, ignoring XML validity, and changing the parsing rules for XML.

Compressing and Squeezing XML

One obvious approach to optimizing XML is to compress it. Since XML is a text-based format, using common binary compression formats like gzip can squeeze over 90% of the volume out of XML data files. However, the problem with compression is that both ends of the communication pipeline must understand the compression format and be able to uncompress the document on the fly without introducing extra latency. A straightforward alternative to using binary compression formats is to simply avoid long element names. Many developers have resorted to the tactic of referring to their XML elements as simply “<g>” or “<bx1>”. While such short tags are definitely an improvement over tags like “<SOAPInspectionSecurityHandler>”, the resulting XML is for all practical purposes no longer human readable.

Another more subtle approach to XML compression is the all-attribute approach. Rather than creating long, complex trees of elements, developers can use a small number of elements with a large number of attributes to dramatically reduce document size. However, the all-attribute approach does not work for complex tree structures, especially those with highly repeating elements. Even with all these creative tricks for reducing the size of XML messages, compression really only serves to reduce network bandwidth utilization and storage requirements, and doesn’t positively impact XML processing performance. In fact, such compression techniques might even reduce the performance.

Ignoring XML Validity

Simply skipping the processing step of validating XML documents is another approach to improving XML performance. In fact, ZapThink’s research has shown that few businesses use XML validation of any type (either DTD or W3C XML Schema) as part of runtime XML-based business processes. Instead, developers will check their XML for validity only during the test or design phases of an implementation, and then simply trust that the documents are remain valid thereafter. After all, checking for a document’s validity does not remove the need to check its validity at the application level anyway. Since validity checking slows down XML parsers, it’s often the first thing to go when optimizing XML performance.

Rewriting the Parser – and Changing the Rules

Since compression and simplified validity checking only minimally improve overall XML processing, developers are increasingly using more drastic approaches to improving XML processing performance. One drastic approach is to rewrite the rules that XML parsers follow. The XML specification is relatively simple – its W3C specification is 80 pages long, which is quite brief considering the power and flexibility of the language. Even at that length, much the XML specification is generally superfluous. When was the last time you used ENTITY, NOTATION, or CDATA elements, anyway? Maybe once in a while those elements come in handy, but they usually aren’t necessary. As a result, developers can recompile their XML so that it contains only a subset of all available XML functionality, which can dramatically increase parser performance by as much as a factor of three.

Yet, the drastic measures to improve XML parsing don’t stop there. Some developers are taking an even more dangerous approach and rewriting the rules of XML itself. For example, some developers eliminate the need for end tags or remove case-sensitivity within XML documents. These approaches are potentially very dangerous moves that sacrifice interoperability and developer-friendliness for speed. In fact, in essence, these developers are creating new, proprietary markup languages of their own.

The ZapThink take

There is no doubt that standard XML is an inefficient data representation format, and the increasing layers of complexity that Web Services add to the core XML language further exacerbate the problem. Nevertheless, ZapThink sees that the many of the above trends for optimizing XML performance are a tremendous step backwards for interoperability and standards-based computing. At what point does a compressed, stripped-down, non-validating “XML-like” format leave the standards behind and represent a proprietary data format? Keep in mind that all of the tactics discussed above require that both ends of a given communication path agree on the optimization mechanism. As a result, the loosely-coupled, implementation-agnostic XML format becomes a tightly-coupled, proprietary implementation, and virtually all of the advantages of using XML for system-to-system communication are lost. ZapThink, therefore, cannot recommend any of the techniques described above.

That being said, we realize that something must be done about improving overall XML performance. The solution is relatively straightforward – more standards. As a set of requirements emerge for optimizing XML performance, so too will a set of agreed-upon conventions and standards for implementing them. Perhaps the WS-I will recommend a particular compression scheme or minimal set of parsing requirements in order to assure interoperability among parties. Maybe a standards body like OASIS or W3C will start work on a new “WS-Compression” specification. Another approach to the problem of XML inefficiency is to leave optimization up to the implementation – basically, trust the application server, XML Proxy, or Web Services management platform to perform the required optimization while leaving the endpoints fully compliant with XML. Companies will then achieve the best of both worlds – high performance without compromising interoperability.