 Customer: IBM Research
Author: Emmanuel Yashchin
Country: United States of America Personally, I consider APL2 the most technically challenging software produced by IBM - ever. Of course, I saw quite a few complex and massive software projects - and I fully appreciate how difficult it is to put together something like MVS or Notes; however, I can imagine how a well funded group of reasonably smart people could make it happen. To produce APL2, however, one would need to have some super-smart people on board... Somehow, this software just "feels right". One of the most important aspects of APL2 is its multi-platform nature. This enables me to write industry-grade applications for use in a mainframe or multi-processor Unix environment by using my preferred Windows development environment, where I can use APL2 in conjunction with so many "goodies" provided by Microsoft and its followers. It is in this environment that I keep my testing setups, logbooks, notes and papers. Once the development is complete, I simply FTP the workspace to a production Unix machine and re-run the tests... just in case (usually, no modifications are needed because I use a set of functions that "adapt" to the operating environment). Then I simply make the workspace available to a production system. This is how APL2 is currently running a massive monitoring program for detection of unfavorable changes in our semiconductor manufacturing operation. This system is in a 24/7 operation for about 1/2 year - and so far, we had 100% availability and a very positive feedback from the users. Most of them do not even know that APL2 is involved in the process. Another recent application (Early Detection Tool) was developed for massive monitoring of IBM Pc Co warranty data. This application was also developed and tested on Windows, but then it was deployed in a mainframe environment prevalent in this division (now Lenovo). The main challenge there was to make sure that the system can explore about 50000 various combinations of machine types and components so as to detect whether some combinations should be brought to the attention of Quality, Procurement or Brands people. The challenging point in such a system is to assure not only good detection capability, but also a low rate of false alarms - and this, in turn, requires some advanced statistical algorithms (these were also developing under APL2). This work led to Research Division Accomplishment in 2004. It is interesting to note that the application was originally developed as a prototype, and it was anticipated that in a subsequent stage this prototype would be re-written into C. However, when the PC Co. people saw that their worldwide warranty data could be fully and reliably processed within 36 hours by just three parallel accounts on the mainframe - they decided to forgo the re-writing and simply develop the system as an APL2 solution. With best regards,
Emmanuel | 
