Self-aware Memory (SaM)
Acceleration of a shallow water simulation for operational application
Self-Organizing and Self-Optimizing Many-Core Architectures
GCC for Transactional Memory
|The DFG-funded project "TM-Opt" researches methods and strategies to analyze, rate, and optimize the runtime behavior of Transactional Memory applications. The analysis aims to uncover conflicting accesses from transactions executed by concurrently running threads. The gathered information is used in an optimization phase to optimize the conflict potential of competing transactions in order to improve the runtime behavior. The research project complements the state of the art in Transactional Memory research.|
Self-aware Memory (SaM) is a decentralized and autonomously
self-optimizing memory management system for scalable many core
architectures with high dynamic application scenarios, in order to
increase the overall system flexibility, dependability and scalability.
Research is done on scalable and dynamic allocation of private and
shared memory, efficient decentralized synchronization techniques,
transactional memory support and especially on autonomous memory
self-optimization e.g. locality optimization.
In addition to the memory management, a decentralized management for allocation of compute resources is investigated.
Exploiting heterogeneous parallel systems poses a new challenge for application
developers. Due to the diversity of such systems, statically choosing
processing units for compute kernel execution can cause low performance and
high energy consumption or even inhibit the execution in case the required
unit is not present.
In this project, we research light-weight concepts and online-learning mechanisms that autonomously analyze the system environment, including competing applications, and adapt execution accordingly during application runtime in order to increase application and overall system performance, dependability and power efficiency.
For disaster management, an existing and verified shallow water simulation is
used to predict the flow of water in case of dam breaks or floodings. As, in
case of an emergency, the results of such a simulation are required as fast
as possible, the usage of high-performance clusters is not feasible as they
cannot be used on demand.
In order to receive the results in reasonable time on available commodity hardware, the goal of this project is to research mechanisms that automatically accelerate the simulation using: a) dynamic regional simplification of the numerical equations and b) effective exploitation of modern heterogeneous parallel systems.
This research project investigates the usage of self-organizing or Organic
Computing principles within dynamically reconfigurable many-core architectures.
Goal of this project is hiding the complexity of such architectures to the user
and easing management and efficient utilization. By using the novel Digital
on-Demand Computing Organism (DodOrg) as evaluation platform, research in this
project covers all areas of self-organizing systems, ranging from system
monitoring up to the realization of a self-optimizing and proactive system
The DodOrg project is a joint research project and is pursued by 4 cooperating chairs from 3 institutes. It is founded through the DFG Priority Program 1183 "Organic Computing".
|The simplified synchronization with Transactional Memory depends on the availability of a compiler that supports TM. For a widespread acceptance and usage of TM, a free and platform independent compiler is mandatory. This gap is addressed by a collaboration in the context of the European Network of Excellence on High Performance and Embedded Architecture and Compilation HiPEAC with Prof. Albert Cohen (INRIA Saclay, France). This collaboration aims at a robust and stable implementation of support for TM in the GCC compiler suite.|