Instant quality control of large batch processing jobs

The most common way to identify success or failure of a job running in a batch-processing mode is by examining a completion code sent by the job to the host operating system. Yet, for a variety of reasons the completion code may inaccurately indicate a successful termination of the job. This article describes a different approach to monitoring the quality of batch processing jobs while in operation. A pattern of behavior is suggested for a program. The pattern reflects ratios of consumption of various hardware resources. The ratios are determined by collecting historical performance variables of the job and analyzing the data by means of statistical methods. Once a pattern is set, the performance variables of every individual run of the program are compared with the precalculated pattern of behavior and if the deviation is beyond certain limits an alarm is triggered. The proposed quality control technique has been tested on real applications, as well as on some artificial programs. The findings suggest that the technique is reliable in that it successfully distinguishes between proper and malfunctioning runs of a program.