Enabling Performance Engineering in Hesse and Rhineland-Palatinate (EPE)
Introduction
Objective: Expand and deepen HPC support in areas where existing scientific expertise coincides with critical user needs.
Approach: Bundle the distributed expertise for HPC support and performance engineering within the HKHLR and the AHRP under a new umbrella to support scientists to efficiently use the HPC resources in Hesse and Rhineland-Palatinate.
Methods
Performance Engineering Services:
Within our project, we design and provide the following portfolio of consulting services:
Scalability
• Objective: Help developers identify and resolve scalability limitations in their codes
• Approach: Define services with increasing degrees of applications engagement around Extra-P, an automatic performance-modeling tool developed at TU Darmstadt
Stability, Performance, and Reproducibility
• Objective: Help developers understand the impact of changes to specific parts of algorithms,especially the exchange of libraries and architectures.
• Approach: Measure condition numbers of algorithms with the help of algorithmic differentiation, by leveraging the tools Code Differentiation Package (CoDiPack) and Message Differentiation Package (MeDiPack) of TU Kaiserslautern. Package run-time environments either in containers or virtual machines.
Multi-/Manycore Performance
• Objective: Help developers to use multi-/manycore architectures more effectively
• Approach: Identify applications that make intensive use of multi-/manycore architectures and suggest optimizations while maintaining portability
Support Structures
• Provide local support staff as point of contact for the above services
• Organize workshops, tutorials, and coding weeks for interested HPC users in Hesse and Rhineland-Palatinate
