18.2 Client/Server based Statistical Computing

While the power of computing equipment increased exponentially, statistical algorithms yielded higher efficiency in terms of computing time and accuracy. Meanwhile the development of high capacity network architectures had another positive impact on this trend, especially the establishment of the world wide web. Consequently a vast number of researchers and programmers in computational statistics as well as institutions like commercial banks, insurers and corporations spent much effort to utilize this evolution for their field of research. An outcome has been the technological philosophy of client/server based statistical computing: meaning a decentralized combination of methods, users and providers of statistical knowledge.

Our understanding of client/server based statistical computing is such that there exists a formal relationship between user, provider and vendor of statistical methodology. An easy to grasp example is a telephone call. The caller (in our case the user demanding statistical methods and/or advice) calls (connects via TCP/IP enabled networks) someone (a high-performance server/vendor of statistical information and methods) who serves his call (the requested calculation is done/information is displayed in a HTML browser, etc.). This client/server understanding is an approach to gain scalability of computational tasks, resource shifting of processing power and decentralization of methods.

There are numerous ways of implementing client/server based statistics, among others Common Gateway Interfaces (CGI), JavaScript, Java Applets and Plug-Ins are the most commonly used techniques. The technology behind the XploRe client/server architecture is thoroughly explained in Kleinow and Lehmann (2002). While that solution is applet based the spreadsheet client presented here has an Add-in character. The M D *ReX -Client is a software tool, nested within an spreadsheet application. In both cases the communication technique relies on the protocol stack M D *CRYPT . For the spreadsheet solution the Java based M D *CRYPT has to be modified. As Microsoft does not support any Java natively for its Office suite, M D *CRYPT has been implemented as a dynamic link library to utilize its interface for Office applications like Excel. The technical aspects and the design philosophy of the M D *ReX -Client are discussed in detail in Aydinli et al. (2002).