WELCOME TO RSDHA 2023!

The RSDHA 2023 workshop is scheduled to take place on November 12, 2023 and will be held at Denver, CO, USA, in conjunction with SC'23.

Official SC RSDHA 2023 session webpage

"Scalable computing" has governed another dimension. Contrary to the traditional definition, which is often proportional to the total number of nodes or transistors working collaboratively to compute a given workload, the newly rising dimension rather scales with the number of different type of processors sharing the computational load. The far end of this dimension is often utilized in energy and latency sensitive environments (e.g. mobile, autonomous and edge computing) where heterogeneity excels most. For example, a state-of-the-art mobile processor such as Apple's A11 series embeds more than 40 specialized IP blocks to pack a very high OPs/Watt ratio in a pocket-portable form factor. On the other hand, node uniformity and processor homogeneity still remain a common trend in high performance computing (HPC) and data center environments, due to simplicity they provide in terms of programmability and scalability.

The proposed workshop targets to investigate an un-explored region of the two-dimensional space of traditional node-based (i.e., horizontal) and heterogeneity-based (i.e., vertical) scaling. Nodes in distributed machines are expected to gradually employ a more diverse set of accelerators to accommodate the increasing computational and power demands of rapidly evolving scientific, machine-learning and data-center workloads. In achieving this goal, there are many lessons to be learned from today's system-on-chips (SoC), such as Apple's M1 and A1X Bionic, and Qualcomm's Snapdragon series, as they already successfully solved aforementioned challenges in the micro-scale. On the other hand, mobile and autonomous platforms embedding these heterogeneous SoCs are becoming more connected to each other (e.g., swarm-computing) and to the cloud (e.g., edge-computing). The techniques used in traditional scaling for HPC could be adopted to overcome the connectivity challenges and enable distributed processing for such systems.

In summary, the proposed workshop seeks answers for two primary questions:

• Vertical Scaling: How could the traditional HPC applications adopt the architectural, programming and runtime approaches employed by the state-of-the-art diversely heterogeneous systems?
• Horizontal Scaling: How could the diversely heterogeneous architectures for mobile and autonomous systems take examples from traditional HPC to beat the multi-node scalability challenges as they become increasingly more connected?

RSDHA 2023 is held in conjunction with SC23: The International Conference for High Performance Computing, Networking, Storage and Analysis.

Workshop Scope (Areas of interest)

The workshop will categorize the panelist expertise and paper submissions under two primary tracks:

• Track 1 - Vertical Scaling: This track will cover the studies that will enable the HPC applications to utilize a more diverse range of processors. The areas of interest include but not limited to:
  • Performance-portable parallel programming paradigms for newly emerging accelerator types.
  • Runtimes for arbitrary degrees of heterogeneous diversity.
  • Analytical modeling for general-purpose and domain-specific programmable accelerators.
  • Application of heterogeneous programming/scheduling techniques used in embedded and mobile computing to HPC systems and vice versa.
  • Case studies using the vertical scaling techniques
• Track 2 - Horizontal Scaling: This track will address the challenges as current highly diverse platforms (e.g., mobile and embedded platforms) are scaled vertically (e.g., teamed up) for collaborative execution. The areas of interest include but not limited to:
  • Communication and memory access challenges for clustered execution of highly diverse platforms.
  • Analytical resource consumption modeling and decision making mechanisms for edge-cloud systems.
  • Compositional resource management and workload balancing across the collaborative clusters.
  • Fault tolerant and resilient execution for mobile and autonomous systems.
  • Case studies using the horizantal scaling techniques

Contact

If you have any problems or questions, please contact us via e-mail at:
belviranli@mines.edu