Propozycja zgłoszona do Ideal-ist Partner Search i opatrzona Quality Label.

PROJECT OVERVIEW
Call Identifier: H2020-FETHPC-2014
Objective: FETHPC 1 – 2014: HPC CORE TECHNOLOGIES, PROGRAMMING ENVIRONMENTS AND ALGORITHMS FOR EXTREME PARALLELISM AND EXTREME DATA APPLICATIONS
Funding Schemes: Research and Innovation Actions
Evaluation Scheme: one stage
Closure Date: 25/11/2014

PROJECT DESCRIPTION

Partner search for FET HPC-1 c) “APIs and system software for future extreme scale systems” to support extreme scale C/C++ application development for future exascale supercomputers.

Proposal Outline:

We develop new APIs and corresponding efficient, flexible and scalable exascale system software for managing extreme scale systems, taking into account extreme parallelism, extreme data, energy consumption and resilience.  A unified programming model will be provided that transparently utilizes available exascale computing resources to achieve unprecedented levels of scalability. The runtime system focuses on overcoming several main barriers to scalability:

• starvation – insufficient concurrent work available to maintain high utilization of all resources

• latencies – the time-distance delay intrinsic to accessing remote resources and services

• overhead – the work required for the management of parallel actions and resources on the critical execution path which is not necessary in a sequential variant

• waiting for contention resolution – the delay due to the lack of availability of oversubscribed shared resources.

In order to evaluate our overall system, we require C/C++ applications that are extremely computational intensive with extreme parallelism and strong potential for nested parallelism and distributed memory architectures.

PARTNER PROFILE SOUGHT

Required skills and Expertise:

We are looking for partners from industry, SME or academia that can provide

extremely computational intensive C/C++ applications with clear potential for extreme scale parallelism to evaluate the project’s programming model and runtime system.

For the proposed project we envision applications that must fulfill the following criteria:

• applications written in C++ or C, AND
• extremely computational intensive with extreme parallelism, AND
• strong potential for nested parallelism and distributed memory parallelization
• compute and data intensive parts must be part of the developer code — not part of some external library or third party code that cannot be restructured at the source code level AND
• MAY exhibit a dynamic, data dependent amount of parallelism.

Applications must have a clear potential for exascale parallel architectures with at least billions of parallel tasks. These applications must already exist in parallel or serial form and must have substantial impact (large user communities or high sales volume). As part of the project their parallel structure will be re-evaluated and if necessary re-structured to reach the granularity and flexibility required for large scale, heterogeneous parallel architectures.

Examples of classes of algorithms which fit our requirements are:

• Branch and bound algorithms

• Place and Route algorithms

• SAT or SMT Solvers, Datalog Engines or similar reasoners

• Combinatorial optimization problems

• Stencil codes, in any number of dimensions

• Dense algebra operations

• N-body simulation

• Rendering.

Description of work to be carried out by the partner(s) sought:

As part of the project, the application developers will have to:

• isolate performance critical code sections and provide them as stand-alone benchmark codes which will be re-structured to achieve exascale performance;
• provide various sets of input and output data to verify the correctness and evaluate the performance of restructured and tuned codes for a wide range of problem sizes; ideally problem instances can be generated for a given problem size;
• free parallel programs of previous parallelization attempts utilizing MPI, OpenCL, CUDA or similar APIs to enable the proper integration of the utilities to be developed by other project partners; Alternatively the application developers may consider a new, parallel implementation utilizing existing knowledge regarding the processed problem and associated algorithms in combination with the utilities provided by the project partners;
•  structure or restructure their application to utilize thread safe data structures as well as data structures suitable for distributed memory architectures.

Type of partner(s) sought:

We are looking for high-profile partners from the EU with extensive EU project experience from

• large industry
• SME
• academia.

Applications must have a large user community or otherwise strong impact (e.g. high sales volume).

PROPOSER INFORMATION
Organisation: University of Innsbruck
Department: Institute of Computer Science
Type of organisation: University
Country: Austria

Więcej…