D4.1: Software for the dynamic adaptivity for patterned applications on initial pattern set for homogeneous hardware architectures.
This document is the initial deliverable for WP4 "Dynamic Adaptation of Parallel Software". The purpose of this work package, according to the DoW, is to develop i) techniques for static mapping of components and data of patterned ap- plications to the available hardware; ii) mechanisms for online machine-learning based scheduling for patterned applications; iii) an adaptive, Just-in-Time (JIT) compilation mechanism for patterned applications; and, iv) infrastructure for monitoring performance of patterned applications. The deliverable is the result of the first phases of tasks T4.1 ("Static Mapping of Software Components and Data to Hardware Resources"), T4.2 ("Adaptive Compilation of Patterned Applications"), T4.3 ("Dynamic Scheduling of Patterned Applications") and T4.4 ("Performance Monitoring of Patterned Applications"). In this deliverable, we give the overview of the basic versions of the tools for static mapping, dynamic scheduling and performance monitoring, together with the JIT compiler.
| d4-1.pdf |
D4.2: Software for the Adaptivity for Patterned Applications on Initial Pattern Set for Heterogeneous Hardware Architectures
This document is the second deliverable for WP4 “Dynamic Adaptation of Parallel Software”. The purpose of this work package, according to the DoW, is to develop i) techniques for static mapping of components and data of patterned applications to the available hardware; ii) mechanisms for online machine-learning based scheduling for patterned applications; iii) an adaptive, Just-in-Time (JIT) compilation mechanism for patterned applications; and, iv) infrastructure for monitoring performance of patterned applications. The deliverable is the result of the first and second phases of tasks T4.1 (“Static Mapping of Software Components and Data to Hardware Resources”), T4.2 (“Adaptive Compilation of Patterned Applications”), T4.3 (“Dynamic Scheduling of Patterned Applications”) and T4.4 (“Performance Monitoring of Patterned Applications”). In this deliverable, we present the ex- tensions to the basic versions of the tools for static mapping, dynamic scheduling and performance monitoring that were described in D4.1. These extensions deal with heterogeneity in the computing environments (for static mapping and performance monitoring) and allow the dynamic scheduling infrastructure to target multiple types of resources and adaptation strategies.
| d4-2.pdf |
D4.3: Software For The Adaptivity For Patterned Applications On Advanced Pattern Set For Heterogeneous Hardware Architectures
This document is the third deliverable for WP4 “Dynamic Adaptation of Parallel Software”. The purpose of this work package, according to the DoW, is to develop i) techniques for static mapping of components and data of patterned applications to the available hardware; ii) mechanisms for online machine-learning based scheduling for patterned applications; iii) an adaptive, Just-in-Time (JIT) compilation mechanism for patterned applications; and, iv) infrastructure for monitoring performance of patterned applications. This deliverable contains the description of the software for the final version of the adaptivity toolset, including static mapping, dynamic scheduling and performance monitoring tools. These tools are described in more detail in D4.4. Here, we provide basic installation and usage instructions for the tools, as well as the locations where the tools can be downloaded from.
| d4-3.pdf |
D4.4: Combined Report On The Final Adaptivity System For Patterned Applications
This document is the fourth deliverable for WP4 “Dynamic Adaptation of Parallel Software”. The purpose of this work package, according to the DoW, is to develop i) techniques for static mapping of components and data of patterned applications to the available hardware; ii) mechanisms for online machine-learning based scheduling for patterned applications; iii) an adaptive, Just-in-Time (JIT) compilation mechanism for patterned applications; and, iv) infrastructure for monitoring performance of patterned applications. The deliverable is the result of the first and second phases of tasks T4.1 (“Static Mapping of Software Components and Data to Hardware Resources”), T4.2 (“Adaptive Compilation of Patterned Ap- plications”), T4.3 (“Dynamic Scheduling of Patterned Applications”) and T4.4 (“Performance Monitoring of Patterned Applications”). In this deliverable, we present the final version of the adaptivity toolset, including static mapping, dynamic scheduling and performance monitoring tools. The basic and intermediate versions of the tools were described in D4.1 and D4.2 respectively. The final ex- tensions of these tools address heterogeneity in the computing environment, both in terms of the type of available processors (CPUs/GPUs) and in terms of scheduling on non-uniform memory architectures (NUMA).
| d4-4.pdf |
THE REPHRASE PROJECT
Copyright © 2018
