XML Pipelining for Mathematical Computation

Exposing a single computation on the Internet via a web service can be challenging. To do so, the input and output of such a service needs to be codified, typically in an XML Schema, and code needs to be written to translate to and from XML. Further, computations need to be interfaced and run with their results formulated into XML. To complicate matters further, there is the whole process of packaging such a system for deployment of some web services platform. While many web services frameworks focus on the interfaces, what is at the core of exposing a web service is the processing of an XML vocabulary (the request elements). Each vocabulary is associated with a set of actions–implicitly or explicitly–to make a wanted computation to take place. Unfortunately, XML isn’t always that simple to process and usually this involves writing a ”pile of code”. This forces the development of web services to be only in the domain of the expert programmer and not that of a Research Mathematician. Further, what then happens when this service needs to deployed into other environments that aren’t necessary a ”web service”? The computation has the same characteristics of input and output, but the packaging and connections are different. Such services should be packaged just as XML services and their ”web services” versions should just be bindings of that to web protocols. In many ways, these problems are very similar to the description of the ”Programming Challenge” for computational sciences as described by Douglas Post [16]. He describes a very similar challenge in writing massive parallel algorithms for grid computing where the desire to build a system is overwhelmed by the programming necessary to implement the system. Certainly, XML-based web services can be part of such grid systems yet very little attention has been paid to the challenges of processing XML or packaging XML services. For XML, there are multitudes of XML-related standards for representing information and for processing XML to accomplish a single task. Many parts of implementing a web service involve using these technologies for specific purposes. Further, any web service request can be reduced to a sequence of ”application steps” where some of the steps are XML technologies like XSLT[5] for transformation, XML Schema [19] for validation, XQuery [2] for queries, etc. But there are no standards for how these technologies are orchestrated into an XML process. This paper proposes utilizing a concept called XML Pipelines for implementing XML-based mathematical computations. In short, an XML pipeline provides a way to chain together the processing steps and, by doing so, simplifies the integration of XML technologies with non-XML components. Ultimately, utilizing an XML pipeline separates concerns and encapsulates complexity–which reduces necessary skill level of the user. They also facilitate deploying XML processes as web services. This paper is distinctively not about SOAP[7], WSDL[4], and the myriad of other web service ”standards.” Those so-called standards are about bindings to protocols on the outside of a service. The technique described here is about the processing that happens inside an XML service. That is, when you start building your XML service, the focus of this paper is the steps and techniques that let you actual implement the processing and connect disparate technologies.