XML and New CPUs?

We have discussed how XML is becoming a dominant description and interface "language." The first step in processing XML statements is to compile XML into native computer bits (machine language?)a process that is CPU intensive (as well as memory and storage intensive). Interestingly, XML processing tends to be stream oriented, more so than other data sources. That is, there is a massive stream of XML syntax pouring into our CPUs, which are presently ill-equipped to turn that stream into another stream of computer bits. Might this need to convert XML streams into data streams that conventional CPUs can interpret lead to a new generation of XML-processing CPUs?

We are already seeing some evidence of this. One company, DataPower, makes a transparent XML pre-processing device that sits in the LAN between servers and watches for XML transactions. Their specialty device can decompose XML at wire speed (that is, it introduces no slowness or additional latency into the data stream) and provides a number of value-add features such as assurance that the syntax is correct (which is an expensive operation, processing-wise), encryption and decryption of the data stream for security, and other tasks. Without such a device, many enterprise applications would need the processing power of several to a dozen servers to match the same performance level and hence the value of a specialty network-based appliance to convert XML streams "on-the-fly."

Today, XML is primarily used on the "boundaries" of a computing network, the points at which a given computer system interfaces to another computer system. XML streams must be converted to another programming language so that they can be interpreted by today's conventional processors. It is likely that, over time, we will see more and more data streams remain in XML format throughout processing the processing steps (i.e., within a processor, between systems, and even stored on disk). This will be done to relieve the processing-overhead impact as XML streams are translated and untranslated multiple times.

A number of companies are chasing the XML chip opportunity, and a number of different approaches will emerge. Initially, these will all appear as niche market opportunities as opposed to the next Intel-class mainstream processor contender. It may be a while (if ever) for the market opportunity to develop to the point that an AMD, Intel, or an IBM/Motorola will build an XML main processor. In today's world, workstations and laptops are our personal bulk data processing devices. In the Inescapable Data world of tomorrow, PDAs and full-function cell phones or gaming handhelds will be the processing devices of choice out on the "edge." If wireless telecommunications take on more of an XML flavor, we can easily imagine that the DSPs (digital signal processors used in all cell phones) of today will be altered to be natural XML-streaming processors.

XML processing has the potential to upset the apple cart regarding the future of mainline networking devices and today's prevalent CPU architectures. Today, the typical LANs or SANs (storage networks) pass data back and forth between servers and storage devices. They are passive networks in that their switching and routing components have extremely little knowledge of the data passing through them. As such, they offer few capabilities for optimizing relationships between data streams and managing of those data streams. If the data elements of "packets" were encapsulated in XML packages, a far higher utility could arise from LAN and SAN components. XML could be used to describe the data inside each packet, thereby making the associations between data packets and servers, databases, and other data sources, for example, more obvious to switches and routers. Therefore, it seems sensible to assume that the encapsulation of data in XML packages is forthcoming, if we can simultaneously drive the performance of XML processors to the point where there is no discernable loss in network performance versus what users now experience. As suggested, it may not be the computer industry itself that drives XML encapsulation; it may be more driven by the communications industry, and then re-absorbed back into mainstream computing.