Some four years ago, AT&T Co got very excited about an algorithm dreamed up by a young scientist of Indian extraction, Narendra Karmarkar, for solving the Travelling Salesman problem. The most striking feature of the problem is its name: it simply involves plotting the most cost- and time-effective route for him to take to make a multitude of calls around the country, but with enough variables, it becomes fiendishly difficult. AT&T itself uses the algorithm to solve sums such as best routing of phone calls; and is now offering the system, called Korbx, including a parallel silicon implementation of the algorithm with a Unix system front end, at $8m to $9m. It has also formed a new Advanced Decision Support Systems division to market it.