The workshop is scheduled virtually on June 21, 2021. Times are EDT (UTC - 4:00). Connection link available upon registration. Workshop start time : 6am PDT, 8am CDT, 9am EDT (USA), 3pm CEST (France, Sweden, Netherlands), 10pm JST (Japan).

The workshop recording is available here. Passcode: j*54wb?%

• 09:00 - 09:10 Opening

• 09:10 - 10:10 Keynote

  • Kate Keahey (University of Chicago / Argonne National Laboratory): Taking Science from Cloud to Edge

• 10:10 - 10:30 Technical Paper Presentation

  • Quincy Wofford (Los Alamos National Laboratory), Patrick Bridges (University of New Mexico), Patrick Widener (Sandia National Laboratory): A Layered Approach for Modular Container Construction and Orchestration in HPC Environments

• 10:30 - 10:50 Break

• 10:50 - 11:50 eScience Invited Talks

  • Christoph Kessler (Linköping University): Portable high-level programming of heterogeneous parallel systems with SkePU

  • Amit Chourasia (San Diego Supercomputing Center): Democratizing Scientific Data Management with SeedMeLab

  • Ben van Werkhoven (eScience Center): GPU code optimization and auto-tuning made easy with Kernel Tuner

  • Eric Coulter (Indiana University): HPC From the Ground to the Cloud

• 11:50 - 12:00 Closing

Christoph Kessler: Portable high-level programming of heterogeneous parallel systems with SkeP

Abstract:

We live in the era of parallel and heterogeneous computer systems, with multi- and many-core CPUs, GPUs and other types of accelerators being omnipresent. The execution and programming models exposed by modern computer architectures are diverse, parallel, heterogeneous, distributed, and far away from the sequential von-Neumann model of the early days of computing. Yet, the convenience of single-threaded programming, together with technical debt from legacy code, make us mentally stick to programming interfaces that follow the familiar von-Neumann model, typically extended with various platform-specific APIs that allow to explicitly control parallelism and accelerator usage. High-level parallel programming based on generic, portable programming constructs known as algorithmic skeletons can raise the level of abstraction and bridge the semantic gap between a sequential-looking, platform-independent single-source program code and the heterogeneous and parallel hardware. We present the design principles of one such framework, the latest generation of our open-source programming framework SkePU for heterogeneous parallel systems and clusters. The SkePU high-level programming interface is based on modern C++, leveraging variadic template metaprogramming and a custom source-to-source pre-compiler. SkePU provides currently seven (fully variadic) skeletons for data-parallel patterns such as map, reduce, stencil, scan etc., together with high-level data abstractions for skeleton call operands. SkePU can perform automated optimizations of the high-level execution flow, such as context-dependent best-backend selection among the supported platforms, operand data transfer and memory optimizations.

Amit Chourasia: Democratizing Scientific Data Management with SeedMeLab

Abstract:

Researchers have an increasing need to manage preliminary and transient results. This evolving and increasing corpus of data needs to be paired with its contextual information and findings. However, this amalgamation of data, metadata, context, and insights is often highly fragmented and dispersed on many systems including local/remote file servers, emails, presentations, and meeting notes. Much of this information becomes increasingly cumbersome to assimilate, use and reuse over time. Researchers often create ad-hoc strategies to manage this data using variety of tools that are loosely glued together, however this fragile system runs into many limitations and burdens them with requiring continued investments. In this talk I will introduce SeedMeLab – an open-source scientific data management system which overcomes limitations of ad-hoc systems by providing a robust feature set that includes data annotation, data discussion, data visualization and discoverability along with modular extensibility. It also provides full ownership, access control and branding; and can be deployable On-Premises or On-Cloud and available as a managed service.

Ben van Werkhoven: GPU code optimization and auto-tuning made easy with Kernel Tuner

Abstract:

Graphics Processing Units (GPUs) have revolutionized the computing landscape in the past decade, and are seen as one of enabling factors in recent breakthroughs in Artificial Intelligence. While GPUs are used to enable scientific computing workloads in many fields, including climate modeling, artificial intelligence, and quantum physics, it is actually very hard to unlock to full computational power of the GPU. This is because there are many degrees of freedom in GPU programming, and often there are only a handful of specific combinations of thread block dimensions and other code optimization parameters, like tiling or unrolling factors, that result in dramatically higher performance than other kernel configurations. To obtain such highly-efficient kernels it is often required to search vast and discontinuous search spaces that consist of all possible combinations of values for all tunable parameters, which is infeasible to do by hand. This talk gives a brief introduction of Kernel Tuner, an easy-to-use tool for testing and auto-tuning OpenCL, CUDA, and Fortran, and C kernels with support for many search optimization algorithms that accelerate the tuning process.

Eric Coulter: HPC From the Ground to the Cloud

Abstract:

This talk describes how we've created and used a “Virtual Cluster toolkit”(VC) and “Container Template Library”(CTL) for building elastic, cloud-based HPC resources to support both Science Gateways and new HPC administrators. The VC toolkit has been invaluable for training for new HPC administrators and users in addition to creating production resources for approximately 25 projects, including SeaGrid, UltraScan, and multiple fast-track COVID-19 research projects. This was based on work from the XSEDE Cyberinfrastructure and Resource Integration (XCRI) team and the Cyberinfrastructure Integration Research Center (CIRC). These teams have been addressing the problem of disconnected and unequally distributed research computing on three fronts: by helping under-resourced institutions build XSEDE-like resources (XCRI), by providing container templates to ease installation and increase portability of scientific software (XCRI), and by enabling easy access to compute resources through science gateways (CIRC). The XCRI team provides both toolkits and hands-on consultation to growing institutions. CIRC supports scientific software developers through Science Gateways using the Airavata middleware, in addition to working with gateway providers to develop software and grow their community. Collaboration between the two resulted in the creation of the toolkits, which have proven beneficial to users and researchers at the edges of the cyberinfrastructure community in the US. In particular, the VC and CTL support several needs surrounding scientific software development in areas less served by established community software, such as rapid development cycles, easing deployment to extant HPC systems, and efficient use of limited cloud resources.