Toshihiro Hanawa
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Toshihiro Hanawa has a total of 171 co-authors. Here are the top co-authors by number of shared publications:
Taisuke Boku
University of Tsukuba
32 shared publicationsMitsuhisa Sato
24 shared publicationsHideharu Amano
Keio University, Yokohama, Japan
10 shared publicationsShin'Ichi Miura
9 shared publicationsYuetsu Kodama
9 shared publicationsKazutami Arimoto
6 shared publications
Latest Publications
- 10 October 2026
- Published by Springer Nature
- p. 69-84
Evaluating UT-Helper for Background Work Offloading in OpenMP Applications
- 10 October 2026
- Published by Springer Nature
- p. 69-84
Evaluating UT-Helper for Background Work Offloading in OpenMP Applications
- Open Access
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Adaptive GPU Sharing for Real-time LLM Serving with Best-effort Workloads
Large Language Models (LLMs) are increasingly deployed in latency-sensitive applications, where real-time serving must satisfy stringent service-level objectives (SLOs). However, request intensities fluctuate over time, and under low load LLM services leave a substantial fraction of GPU task idle. A promising approach to reclaim this capacity is co-locating best-effort workloads with LLM serving via GPU sharing. In this study, we propose AdMix, a scheduling system that enables adaptive GPU sharing between real-time LLM serving and best-effort non-LLM deep learning (DL) inference. AdMix dynamically regulates the best-effort workload’s resource allocation using NVIDIA Multi-Process Service (MPS), guided by real-time monitoring of LLM performance and a lightweight latency estimator. By real-world LLM serving traces, experiments on LLaMA-7B and Qwen-14B paired with representative DL workloads demonstrate that AdMix improves best-effort throughput by 1.3–5 × over MPS baseline while preserving LLM SLO attainment.
- 27 September 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-6
Open AccessAdaptive GPU Sharing for Real-time LLM Serving with Best-effort Workloads
Large Language Models (LLMs) are increasingly deployed in latency-sensitive applications, where real-time serving must satisfy stringent service-level objectives (SLOs). However, request intensities fluctuate over time, and under low load LLM services leave a substantial fraction of GPU task idle. A promising approach to reclaim this capacity is co-locating best-effort workloads with LLM serving via GPU sharing. In this study, we propose AdMix, a scheduling system that enables adaptive GPU sharing between real-time LLM serving and best-effort non-LLM deep learning (DL) inference. AdMix dynamically regulates the best-effort workload’s resource allocation using NVIDIA Multi-Process Service (MPS), guided by real-time monitoring of LLM performance and a lightweight latency estimator. By real-world LLM serving traces, experiments on LLaMA-7B and Qwen-14B paired with representative DL workloads demonstrate that AdMix improves best-effort throughput by 1.3–5 × over MPS baseline while preserving LLM SLO attainment.
- Open Access
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 353-360
Evaluating Claude Code’s Coding and Test Automation for GPU Acceleration ofa Legacy Fortran Application: A GeoFEM Case Study
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 353-360
Open AccessEvaluating Claude Code’s Coding and Test Automation for GPU Acceleration ofa Legacy Fortran Application: A GeoFEM Case Study
- Open Access
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 153-164
Tensor-Core-Optimized Strategies for BLR × Tall-Skinny Matrix Multiplication in BEM
- 25 January 2026
- Published by Association for Computing Machinery (ACM)
- p. 153-164
Open AccessTensor-Core-Optimized Strategies for BLR × Tall-Skinny Matrix Multiplication in BEM
- 24 November 2025
- Published by Springer Nature
- p. 271-283
Analysis of Application Power Characteristics Using Performance Counters on A64FX
- 24 November 2025
- Published by Springer Nature
- p. 271-283
Analysis of Application Power Characteristics Using Performance Counters on A64FX
- Open Access
- 19 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Accelerating Heterogeneous Coupling Computing with WaitIO Using RDMA
- 19 February 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Open AccessAccelerating Heterogeneous Coupling Computing with WaitIO Using RDMA
- Open Access
- 29 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 181644-181665
Unified Schemes for Directive-Based GPU Offloading
GPU is the dominant accelerator device due to its high performance and energy efficiency. Directive-based GPU offloading using OpenACC or OpenMP target is a convenient way to port existing codes originally developed for multicore CPUs. Although OpenACC and OpenMP target provide similar features, both methods have pros and cons. OpenACC has better functions and an abundance of documents, but it is virtually for NVIDIA GPUs. OpenMP target supports NVIDIA/AMD/Intel GPUs but has fewer functions than OpenACC. Here, we have developed a header-only library, Solomon (Simple Off-LOading Macros Orchestrating multiple Notations), to unify the interface for GPU offloading with the support of both OpenACC and OpenMP target. Solomon provides three types of notations to reduce users’ implementation and learning costs: intuitive notation for beginners and OpenACC/OpenMP-like notations for experienced developers. This manuscript denotes Solomon’s implementation and usage and demonstrates the GPU-offloading in N-body simulation and the three-dimensional diffusion equation. The library and sample codes are provided as open-source software and publicly and freely available at https://github.com/ymiki-repo/solomon.
- 29 November 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 181644-181665
Open AccessUnified Schemes for Directive-Based GPU Offloading
GPU is the dominant accelerator device due to its high performance and energy efficiency. Directive-based GPU offloading using OpenACC or OpenMP target is a convenient way to port existing codes originally developed for multicore CPUs. Although OpenACC and OpenMP target provide similar features, both methods have pros and cons. OpenACC has better functions and an abundance of documents, but it is virtually for NVIDIA GPUs. OpenMP target supports NVIDIA/AMD/Intel GPUs but has fewer functions than OpenACC. Here, we have developed a header-only library, Solomon (Simple Off-LOading Macros Orchestrating multiple Notations), to unify the interface for GPU offloading with the support of both OpenACC and OpenMP target. Solomon provides three types of notations to reduce users’ implementation and learning costs: intuitive notation for beginners and OpenACC/OpenMP-like notations for experienced developers. This manuscript denotes Solomon’s implementation and usage and demonstrates the GPU-offloading in N-body simulation and the three-dimensional diffusion equation. The library and sample codes are provided as open-source software and publicly and freely available at https://github.com/ymiki-repo/solomon.
- Open Access
- 11 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 77-82
Optimize Efficiency of Utilizing Systems by Dynamic Core Binding
Load balancing at both the process and thread levels is imperative for minimizing application computation time in the context of MPI/OpenMP hybrid parallelization. This necessity arises from the constraint that, within a typical hybrid parallel environment, an identical number of cores is bound to each process. Dynamic Core Binding, however, adjusts the core binding based on the process’s workload, thereby realizing load balancing at the core level. In prior research, we have implemented the DCB library, which has two policies for computation time reduction or power reduction. In this paper, we show that the two policies provided by the DCB library can be used together to achieve both computation time reduction and power consumption reduction.
- 11 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 77-82
Open AccessOptimize Efficiency of Utilizing Systems by Dynamic Core Binding
Load balancing at both the process and thread levels is imperative for minimizing application computation time in the context of MPI/OpenMP hybrid parallelization. This necessity arises from the constraint that, within a typical hybrid parallel environment, an identical number of cores is bound to each process. Dynamic Core Binding, however, adjusts the core binding based on the process’s workload, thereby realizing load balancing at the core level. In prior research, we have implemented the DCB library, which has two policies for computation time reduction or power reduction. In this paper, we show that the two policies provided by the DCB library can be used together to achieve both computation time reduction and power consumption reduction.
- Open Access
- 11 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 63-68
MPI-Adapter2: An Automatic ABI Translation Library Builder for MPI Application Binary Portability
This paper proposes an automatic MPI ABI (Application Binary Interface) translation library builder named MPI-Adapter2. The container-based job environment is becoming widespread in computer centers. However, when a user uses the container image in another computer center, the container with MPI binary may not work because of the difference in the ABI of MPI libraries. The MPI-Adapter2 enables to building of MPI ABI translation libraries automatically from MPI libraries. MPI-Adapter2 can build MPI ABI translation libraries not only between different MPI implementations, such as Open MPI, MPICH, and Intel MPI but also between different versions of MPI implementation. We have implemented and evaluated MPI-Adapter2 among several versions of Intel MPI, MPICH, MVAPICH, and Open MPI using NAS parallel benchmarks and pHEAT-3D, and found that MPI-Adapter2 worked fine except for Open MPI ver. 4 binary on Open MPI ver. 2 on IS of NAS parallel benchmarks, because of the difference in MPI object size. We also evaluated the pHEAT-3D binary compiled by Open MPI ver.5 using MPI-Adapter2 up to 1024 processes with 128 nodes. The performance overhead between MPI-Adapter2 and Intel native evaluation was 1.3%.
- 11 January 2024
- Published by Association for Computing Machinery (ACM)
- p. 63-68
Open AccessMPI-Adapter2: An Automatic ABI Translation Library Builder for MPI Application Binary Portability
This paper proposes an automatic MPI ABI (Application Binary Interface) translation library builder named MPI-Adapter2. The container-based job environment is becoming widespread in computer centers. However, when a user uses the container image in another computer center, the container with MPI binary may not work because of the difference in the ABI of MPI libraries. The MPI-Adapter2 enables to building of MPI ABI translation libraries automatically from MPI libraries. MPI-Adapter2 can build MPI ABI translation libraries not only between different MPI implementations, such as Open MPI, MPICH, and Intel MPI but also between different versions of MPI implementation. We have implemented and evaluated MPI-Adapter2 among several versions of Intel MPI, MPICH, MVAPICH, and Open MPI using NAS parallel benchmarks and pHEAT-3D, and found that MPI-Adapter2 worked fine except for Open MPI ver. 4 binary on Open MPI ver. 2 on IS of NAS parallel benchmarks, because of the difference in MPI object size. We also evaluated the pHEAT-3D binary compiled by Open MPI ver.5 using MPI-Adapter2 up to 1024 processes with 128 nodes. The performance overhead between MPI-Adapter2 and Intel native evaluation was 1.3%.
- Open Access
mdx: A Cloud Platform for Supporting Data Science and Cross-Disciplinary Research Collaborations
The growing amount of data and advances in data science have created a need for a new kind of cloud platform that provides users with flexibility, strong security, and the ability to couple with supercomputers and edge devices through high-performance networks. We have built such a nation-wide cloud platform, called "mdx" to meet this need. The mdx platform's virtualization service, jointly operated by 9 national universities and 2 national research institutes in Japan, launched in 2021, and more features are in development. Currently mdx is used by researchers in a wide variety of domains, including materials informatics, geo-spatial information science, life science, astronomical science, economics, social science, and computer science. This paper provides an the overview of the mdx platform, details the motivation for its development, reports its current status, and outlines its future plans.
Open Accessmdx: A Cloud Platform for Supporting Data Science and Cross-Disciplinary Research Collaborations
The growing amount of data and advances in data science have created a need for a new kind of cloud platform that provides users with flexibility, strong security, and the ability to couple with supercomputers and edge devices through high-performance networks. We have built such a nation-wide cloud platform, called "mdx" to meet this need. The mdx platform's virtualization service, jointly operated by 9 national universities and 2 national research institutes in Japan, launched in 2021, and more features are in development. Currently mdx is used by researchers in a wide variety of domains, including materials informatics, geo-spatial information science, life science, astronomical science, economics, social science, and computer science. This paper provides an the overview of the mdx platform, details the motivation for its development, reports its current status, and outlines its future plans.