Weighted-balanced testing procedure with distributed testers

This talk analyses the performance of the weight-balanced testing algorithm (WBT) with distributed testers. The WBT objective is to minimize the expected number of tests (or round of tests in the case of multiple testers) to find an item in an ordered list/sequence. For example, the algorithm has been suggested for searching a nonconforming part within a manufactured batch or for discovering a defective segment in a communication line. The application areas for the WBT are widespread and include quality control, coding, memory storage and testing applications [1],[2],[3],[4]. In this talk, we show that the WBT obtains a reasonable search time if it is implemented with a single tester. However, when it is implemented with distributed testers, the algorithm often obtains a poor performance. In particular, we demonstrate that the distributed-testers WBT is far from being the optimal search procedure, which is obtained by applying the Hu-Tucker algorithm. This result is important since the performance of a distributed-testers WBT and, particularly, its upper bound on the search time have not been analyzed before, despite a large body of literature that suggest it for various testing applications [2][3]. The talk follows [2] and provides a numerical comparison of several testing algorithms. The performance measures used for the comparison are the Average Search Length (ASL) and the search upper-bounds (UB). The comparison is based on known testing scenarios found in the literature as well as some special cases. Finally, we discus some implementation approaches for multiple testers, such as central versus distributed testing, in-light of the work of Abrahams [1]. Several numerical examples are presented for illustration purposes.