Duties include project management, system administration, benchmark coordination, technical support, quality assurance planning, and miscellaneous technical tasks.
I was hired to establish the use of Linear Programming (LP) and related techniques on Cray hardware. Because I was the only person at Cray whose focus was in this area, my efforts ranged from code and algorithm development to sales strategy and support. Technical achievements include the development, in conjunction with third-party software vendors, of highly vectorized large-scale codes using both simplex and interior point technologies. Starting from a point in 1986 when Cray was unable even to solve large LP benchmarks supplied by prospects, we progressed to the point of being able to claim the fastest solution of LP's in the world. In addition to the mathematically oriented work needed to accomplish this, I had to maintain a close liaison with Cray's compiler developers and performance utility developers in order to assure the availability of necessary tools. Apart from working directly on these aspects of developing Cray's LP capabilities, I was
Starting in 1993, my duties shifted to the area of Structural Analysis, to perform the porting and tuning of LS-DYNA3D, a third-party code, to Cray's massively parallel machine, the CRAY T3D. This code was designated as critical to the success of the T3D, and its successful placement at Chrysler fulfilled the 1994 corporate goal to penetrate the automotive industry with the T3D. The technical goal, of having 32 processors on the T3D equal in speed to a C90 vector processor, was met. Work in afterward centered on making LS-DYNA3D ready to take advantage of the features of the T3D's follow-on product, the T3E.
I joined a small group of LP people at CDC, initially to run their "hot line" telephone/telex support service for their LP solver APEX-III and matrix generator PDS/MAGEN (the latter being a product of Haverly Systems). These products were sold on CDC mainframes, and CDC ran its own timesharing service (CYBERNET) through which APEX and PDS/MAGEN also drew substantial revenues. In addition to on-demand troubleshooting, I developed and taught classes in the use of the products, and did presales and postsales LP modeling. When an enhancement project, APEX-IV, was undertaken in the early 1980's, I participated in each phase (requirements, design, coding, testing). My main areas of responsibility in this two-year period were the command-language processor and the matrix scaling algorithm. I also assisted in the enhancement of the matrix reduction feature of the product. In the mid-1980's I performed the port of SCICONIC and other, non-LP, codes to the NOS/VE operating system on their new Cyber 180 line of hardware. After that, I worked two years on a PC-based front-end for mainframe engineering applications.
In my 18 months with CSC, I worked on a defense contract, under clearance, as part of a three-person effort to enhance the performance of a military logistics simulation code. Keys to the success of this project were the incorporation of existing software the users had confidence in, and the recognition that the underlying problem was combinatorial in nature. Simple LP techniques couldn't give useful answers, and Branch-and-Bound algorithms would have been ineffective for this size of problem. I therefore devised a heuristic that allocated shipments to air or sea in units rather than in pieces, as a natural first step in their simulator code. The users could thus treat our work as an enhancement of their own code, and began using it immediately.