PIs: Vitali Morozov, Kaylan Kumaran, and Venkat Vishwanath, LCF, Prasanna Balaprakash, MCS
Objective: Understanding a workload behaviour has already become a critical step in performance studies; it plays an important role in the design and adoption of new algorithms and hardware. Our objective is to develop a performance analysis, tuning, and projection framework named SKOPE – SKeleton for wOrkload Performance Exploration.
SKOPE provides:
- An abstract language for algorithmic description of a workload.
- A framework for internal representation of the workload.
- An API to describe the target hardware platform.
- A set of tools to perform performance analysis of a workload on a given platform.
- A set of tools to generate performance projection of a workload on a given platform.
- A tool to allow for transformations to better tuning a workload for the hardware.
Testbed: We will be using a diversity of computer architectures and memory subsystems for ALCF workload performance analysis and explorations.