PPoPP ExHET 2024

ExHET 2024

The 3rd International Workshop on Extreme Heterogeneity Solutions

to be held in conjunction with
PPoPP 2024

03 March, 2024
Edinburgh, UK

Introduction

While computing technologies have remained relatively stable for nearly two decades, new architectural features, such as specialized hardware, heterogeneous cores, deep memory hierarchies, and near-memory processing, have emerged as possible solutions to address the concerns of energy-efficiency, manufacturability, and cost. However, we expect this ‘golden age’ of architectural change to lead to extreme heterogeneity and it will have a major impact on software systems and applications. In this upcoming exascale and extreme heterogeneity era, it will be critical to explore new software approaches that will enable us to effectively exploit this diverse hardware to advance science, the next-generation systems with heterogeneous elements will need to accommodate complex workflows. This is mainly due to the many forms of heterogeneous accelerators (no longer just GPU accelerators) in this heterogeneous era, and the need of mapping different parts of an application onto elements most appropriate for that application component.

Objectives, scope and topics of the workshop

The goal of this workshop is to provide a forum to discuss new and emerging solutions to address these important challenges from the upcoming extreme heterogeneity era. Papers are being sought on many aspects of heterogeneous computing including (but not limited to):

Program

1,20 PM - 3,20 PM : Session 1: Chair, Akshay Bhosale
1,20 PM - 1,25 PM : Opening Remarks - Gokcen and Akshay (5 minutes)
1,25 PM - 2,20 PM : Nick Brown Keynote (55 minutes)
2,20 PM - 2,50 PM : Paper 1 (30 minutes) Enhancing Intra-Node GPU-to-GPU Performance in MPI+UCX through Multi-Path Communication
2,50 PM - 3,20 PM : Paper 2 (30 minutes) GPU-Initiated Resource Allocation for Irregular Workloads

3,20 PM - 3,40 PM : Coffee Break

3,40 PM - 5,40 PM : Session 2 Chair, Gokcen Kestor
3,40 PM - 4,10 PM : Hyesoon Kim Invited Talk (30 minutes)
4,10 PM - 4,40 PM : Paper 3 (30 minutes) Preparing for Future Heterogeneous Systems Using Migrating Threads
4,40 PM - 5,30 PM : Panel (50 minutes)
5,30 PM - 5,40 PM : Best Paper Award and Concluding Remarks (10 minutes)

2024 Best Paper Award

Enhancing Intra-Node GPU-to-GPU Performance in MPI+UCX through Multi-Path Communication
Authors: Amirhossein Sojoodi (Queen's University), Yiltan H. Temucin (Queen's University) and Ahmad Afsahi (Queen's University). Abstract: Efficient communication among Graphics Processing Unit (GPU)s is crucial for achieving high performance in modern GPU-accelerated applications. This paper introduces a multi-path communication framework within the MPI+UCX library to enhance Point-to-Point (P2P) communication performance between intra-node GPUs, by concurrently leveraging multiple paths, including available NVLinks and PCIe through the host. Through extensive experiments, we demonstrate the significant performance gains achieved by our approach, surpassing baseline P2P communication methods. More specifically, multi-path P2P can improve the bandwidth of UCX Put operation by up to 2.85x. Furthermore, we demonstrate the effectiveness of our approach in accelerating the Jacobi iterative solver, achieving up to 1.27x runtime speedup.

Important Dates

Paper submission deadline : December 29, 2023
Notification of acceptance : January 12, 2024
Camera-ready papers due : February 26, 2024
Workshop day: March 03, 2024

Steering Committee

Jeffrey S. Vetter, Oak Ridge National Laboratory, USA

Olivier Aumage, INRIA, France

Manuel Prieto, University Complutense of Madrid, Spain

Hartwig Anzt, KIT, Germany

Stanimire Tomov, University of Tennessee at Knoxville, USA

Antonio J. Pena, Barcelona Supercomputing Center, Spain, Spain

Organizers (Contact us)

Pedro Valero-Lara (co-chair)
Oak Ridge National Laboratory, USA
valerolarap@ornl.gov

Seyong Lee (co-chair)
Oak Ridge National Laboratory, USA
lees2@ornl.gov

Gokcen Kestor (co-chair)
Pacific Northwest National Laboratory, University of California Merced, USA
gokcen.kestor@pnnl.gov

Monil Mohammad Alaul Haque (proceeding chair and program chair)
Oak Ridge National Laboratory, USA
monilm@ornl.gov

Jose Manuel Monsalve Diaz (publicity chair)
Argonne National Laboratory, USA
jmonsalvediz@anl.gov

Simon Garcia de Gonzalo (web chair)
Sandia National Laboratory, USA
simgarc@sandia.gov

Programme Committee

Manuscript submission

Papers reporting original and unpublished research results and experience are solicited. Papers must not exceed 6 pages in standard ACM two-column conference format. ACM templates for Microsoft Word, and LaTeX are available here. All paper submissions will be handled electronically via EasyChair.

Proceedings

All accepted papers will be published in the ExHET-PPoPP Workshops 2024 proceedings by the ACM Digital Library.

Best Paper Award

The Best Paper Award will be selected on the basis of explicit recommendations of the reviewers and their scoring towards the paper’s originality and quality.

Special Issue Journal

Selected best papers of ExHET will be considered for publication in a special issue of the international journal Applied Sciences (IF: 2.838).

Keynote (Nick Brown, University of Edinburgh):

Domain-specific abstractions: A panacea or false promise for programming extreme heterogeneous architectures Dr. Nick Brown

There is a saying that it's called hardware because developing it is hard. However, in recent years there has been a Cambrian explosion in novel hardware architectures that one can use to undertake their calculations and the major blocker is now in fact on the software side. HPC developers must be able to effectively exploit such a diverse mix of hardware without significant architecture-specific expertise and extensively modifying their code for each technology. This is something that we are currently unable to provide but it is critical that we solve the issue, not least because such a heterogeneous mix can often deliver very significant energy advantages. In this talk, I will discuss our efforts to leverage knowledge of the programmer's intentions via domain-specific abstractions during compilation using MLIR. In our approach, computational patterns are identified and then drive the compilation process which enables a single, unmodified, source code to target a wide variety of architectures including GPUs, FPGAs, CGRAs, and the CS-2. By building upon the MLIR infrastructure we are able to often deliver a step change in performance and productivity compared to the current state of the art.

Dr. Nick Brown is a Senior Research Fellow at EPCC, University of Edinburgh with research interests in novel hardware architectures and how we empower HPC developers to leverage these most effectively without extensive architecture-specific knowledge. Much of his research has focussed on developing algorithmic techniques to enable the optimization of HPC codes on architectures such as FPGAs and RISC-V, and he runs testbeds as part of the ExCALIBUR exascale program which provides access to these for developers to experiment with their workloads on. He is also the knowledge exchange coordinator for the xDSL project which aims to develop a common DSL compiler ecosystem based on LLVM and MLIR. He has developed numerous production HPC simulation codes in the past, including the Met Office's high-resolution atmospheric model called MONC.

Invited Talk (Hyesoon Kim, Georgia Institute of Technology):

Modeling of Heterogeneous Computing Systems

Heterogeneous computing has become the main computing platform as we are in the realm of accelerators. Traditionally, heterogeneous computing was centered around CPUs. However, with the explosion of required computing power and large data, GPUs have become the primary center of computation, forming a heterogeneous computing platform.One of the biggest challenges of using a heterogeneous computing platform is proper workload partitioning. In this talk, I will discuss several modeling techniques to predict the performance of different heterogeneous computing platforms.

Dr. Hyesoon Kim is a professor in the School of Computer Science at the Georgia Institute of Technology and a co-director of the Center for Novel Computing Hierarchy. Her research areas include the intersection of computer architectures and compilers, with an emphasis on heterogeneous architectures such as GPUs and near-data processing. She is a recipient of the NSF Career Award and is a member of the MICRO/HPCA Hall of Fame. She is the chair of IEEE TCuARCH and an associate editor of Transactions on Architecture and Code Optimization. She is also an IEEE fellow.

Panel:

Challenges and Solutions for the upcoming Extreme Heterogeneity Era

During the panel discussion, the panelists as well as those participants in the workshop, will have the opportunity to discuss the fundamentals of extreme heterogeneity: challenges and solutions.

Panelists:

Nick Brown, University of Edinburgh, UK

Dong Li, University of Californio Merced, USA

Jacques Pienaar, Google, USA

Antonino Tumeo, Pacific Northwest National Laboratory, USA

Registration

Information about registration at PPoPP 2024 website.