Details of Grant 

EPSRC Reference:	GR/K19723/01
Title:	CONTROL DECOUPLING IN HIGH PERFORMANCE COMPUTING SYSTEMS
Principal Investigator:	Topham, Professor N
Other Investigators:	Pooley, Professor R
Researcher Co-Investigators:
Project Partners:
Department:	Sch of Informatics
Organisation:	University of Edinburgh
Scheme:	Standard Research (Pre-FEC)
Starts:	01 October 1994	Ends:	30 September 1997	Value (£):	151,775
EPSRC Research Topic Classifications:	Parallel Computing
EPSRC Industrial Sector Classifications:
Related Grants:
Panel History:
Summary on Grant Application Form
The main objective of this project is to investigate the ways in which decoupled execution can contribute to future high performance computer architectures. We shall concentrate efforts on a new form of decoupling known as control decoupling . The goal of control decoupling is to support run-time evaluation of a variety of program actions; particularly those which enhance execution efficiency.Progress:The first objective the project has been to design, and construct the tools necessary for assessing the benefits of control decoupling. The primary tool being developed is the Dynamic Data Flow Analyser (DDFA). This is a tool which annotates the source code by introducing shadow variables to provide detailed run-time information about the execution of the program. This approach has been used previously to evaluate the effectiveness of dependency tests, to compare speed up for different granularities in parallel architectures, and to evaluate scheduling techniques. It is a powerful technique, and our first goal is therefore to equip ourselves with good compilation tools with which we can apply the technique to a variety of our own problems.Although other researchers have used this technique, we believe we are the first to see the annotated source code as a form of standard iterative framework; when the program is executed it acts as perfect data flow analyser, allowing us to gather information about the characteristics of the program (this will be important for understanding those dependencies where control decoupling may be used).The DDFA is written in C++, allowing us to utilise some of the combinatorial libraries developed by the Max Planck institute. The tool takes a G(V,E) where V are the vertices, and E the edges of the graph, and by generating the appropriate annotation DDFA ensures that the information from the source is taken to the sink vertex; the vertices may be the basic operations, loops, blocks, or procedures. For example when considering the earliest time a variable becomes available, a dependency edge between two vertices enforces a synchronisation. Using the notion of the standard iterative framework, the user of the DDFA need only define a confluence, transfer, and initialisation function to generate the annotation for the data flow analyser they require. This approach will give us sufficient flexibility to perform the following experiments: o Using the decoupling algorithm to evaluate the number loss of decouplings between the control and execution nodes. Using the decoupling algorithms to evaluate the amount of idle time a control node in an abstract perfect decoupled machine. Gather information about the dependencies within the program which can not be resolved by current tests. Evaluate the programs speedup using control decoupling. It is expected that we should be able to perform the first simple experiments with DDFA in the next month.
Key Findings
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Potential use in non-academic contexts
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Impacts
Description This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk Summary Date Materialised
Sectors submitted by the Researcher
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Project URL:
Further Information:
Organisation Website:	http://www.ed.ac.uk