Parallel Computing

Special issue on Hardware/Software Co-design for Sparse and Irregular Applications

develop own website
Mobirise

Call for Papers

Many well established or emerging high-performance computing applications are said to exhibit irregular behaviors, because they present fine-grained unpredictable memory access patterns, irregularity in the contro control structures, and/or network communication of variable sizes. They operate on ever growing, not rigidly structured, data sets, thus they have significant degree of parallelism. However, for these same reasons they may also be synchronization intensive, and very difficult to load balance on parallel architectures. 

Irregular applications pertain both to well established and emerging fields, such as Computer Aided Design (CAD), bioinformatics, data mining, machine learning on sparse data structures (e.g. Graph Neural Networks), analysis of social, transportation, communication and other types of networks, and computer security. Additionally, the newest high-performance computing applications really are converging towards a mix of conventional scientific simulation, machine learning and data analytics, hence combining regular with irregular irregular phases. 

Current high-performance systems rely on data locality, regular computations, and easily partitionable data sets to exploit parallelism and increase performance. This happens both at the hardware level, where general purpose and specialized processors (e.g., GPUs, or domain-specific accelerators mostly focused to machine learning) aim at reaching high flop-rates with vector or tensor units for computation on dense data structures and at reducing latencies with large caches and deep memory hierarchies, and at the software level, where most of the runtimes, libraries, and/or algorithms mostly exploit data partitioning and data movement reduction. However, the current solutions that focus on leveraging these features do not cope well with the more complex sparse data structures and behaviors of irregular applications. 

Addressing the issues of irregular applications on current and future system architectures will become critical to solve the scientific challenges of the next few years, and will require a collaborative codesign process of both the hardware and the software.

This special issue seeks works that explore hardware/software co-design approaches for developing and optimizing execution of irregular applications on high-performance systems, at all levels of the stack: micro- and systemarchitecture, network, languages, libraries, runtimes, compilers, analysis, algorithms. Specifically, this issue aims at collecting those novel research solutions that connect domain-specific architectures and emerging high-performance applications that exhibit irregular behaviors. 

 Topics of interest, of both theoretical and practical significance, include but are not limited to: 

•Micro- and System-architectures 
Network and memory architectures
Manycore, hybrid, and custom architectures (Tensor architectures, GPUs, FPGAs, near-memory designs)
Heterogeneous approaches and methods for exploiting domain-specific systems
Modeling, evaluation and characterization of domain-specific architectures for memory intensive and irregular applications also from theoretical perspective
•Innovative algorithmic techniques
•Parallelization techniques for sparse data structures
Languages and programming models
Library and runtime support
Compiler and analysis techniques
•Case studies of irregular applications (e.g. Knowledge Graphs, Machine Learning on sparse data-structures, Data Mining, Security, Bioinformatics)
Support for irregular applications on novel computing architectures (quantum, neuromorphic)


Important Dates

  • Submissions open: December 1, 2020 
  • Submission Deadline: March 29, 2021 (EXTENDED)
  • Acceptence Notification: September 1, 2021

Submissions

Instructions and templates for prospective author are provided at the link: https://www.elsevier.com/journals/parallel-computing/0167-8191/guide-for-authors

The submission site is accessible from Editorial Manager, at the link: https://www.editorialmanager.com/PARCO/default.aspx

Authors must select article type: “Special Issue: Hardware/Software Co-design for Sparse and Irregular Applications” in Editorial Manager for the submission to be considered for this special issue.


Guest Editors

For any questions or additional information regarding this Special Issue, please contact the Guest Editors
  • Flavio Vella (Free University of Bozen), flavio.vella@unibz.it
  • Antonino Tumeo (PNNL), antonino.tumeo@pnnl.gov