Abstract: A generic performance model is generated for a program job, wherein the program job executes one or more map tasks in a map phase and one or more reduce tasks in a reduce phase in a distributed computing system. The generic performance model is trained to generate a trained performance model based on historical performance of the program job and a set of one or more program job-specific parameters. Performance of a subsequent execution of the program job is predicted based on the trained performance model.

Abstract: Systems and methods for tuning photolithographic processes are described. A model of a target scanner is maintained defining sensitivity of the target scanner with reference to a set of tunable parameters. A differential model represents deviations of the target scanner from the reference. The target scanner may be tuned based on the settings of the reference scanner and the differential model. Performance of a family of related scanners may be characterized relative to the performance of a reference scanner. Differential models may include information such as parametric offsets and other differences that may be used to simulate the difference in imaging behavior.