Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Scalable, Cloud-Based Simulations of Blood Flow and Targeted Drug Delivery in Retinal Capillaries

Lucas Amoudruz, Sergey Litvinov, Riccardo Murri, Volker Eyrich, Jens Zudrop, Costas Bekas, Petros Koumoutsakos

TL;DR

The paper assesses cloud HPC as a viable platform for large-scale, tightly coupled microscale blood-flow simulations. It compares two robust particle-based tools, Mirheo and LAMMPS, across DPD-based plasma, RBC membranes, and ABF propulsion, demonstrating strong weak and strong scaling on public clouds and on a national supercomputer. Key contributions include optimized membrane connectivity handling in LAMMPS, efficient ABF communication strategies, and the first demonstration of ABF dynamics within a retinal capillary network derived from fundus imagery. The findings suggest cloud resources can democratize access to complex, high-fidelity biophysical simulations with performance competitive to traditional HPC, enabling broader, personalized biomedical research.

Abstract

We investigate the capabilities of cloud computing for large-scale,tightly-coupled simulations of biological fluids in complex geometries, traditionally performed in supercomputing centers. We demonstrate scalable and efficient simulations in the public cloud. We perform meso-scale simulations of blood flow in image-reconstructed capillaries, and examine targeted drug delivery by artificial bacterial flagella (ABFs). The simulations deploy dissipative particle dynamics (DPD) with two software frameworks, Mirheo (developed by our team) and LAMMPS. Mirheo exhibits remarkable weak scalability for up to 512 GPUs. Similarly, LAMMPS demonstrated excellent weak scalability for pure solvent as well as for blood suspensions and ABFs in reconstructed retinal capillaries. In particular, LAMMPS maintained weak scaling above 90% on the cloud for up to 2,000 cores. Our findings demonstrate that cloud computing can support tightly coupled, large-scale scientific simulations with competitive performance.

Scalable, Cloud-Based Simulations of Blood Flow and Targeted Drug Delivery in Retinal Capillaries

TL;DR

The paper assesses cloud HPC as a viable platform for large-scale, tightly coupled microscale blood-flow simulations. It compares two robust particle-based tools, Mirheo and LAMMPS, across DPD-based plasma, RBC membranes, and ABF propulsion, demonstrating strong weak and strong scaling on public clouds and on a national supercomputer. Key contributions include optimized membrane connectivity handling in LAMMPS, efficient ABF communication strategies, and the first demonstration of ABF dynamics within a retinal capillary network derived from fundus imagery. The findings suggest cloud resources can democratize access to complex, high-fidelity biophysical simulations with performance competitive to traditional HPC, enabling broader, personalized biomedical research.

Abstract

We investigate the capabilities of cloud computing for large-scale,tightly-coupled simulations of biological fluids in complex geometries, traditionally performed in supercomputing centers. We demonstrate scalable and efficient simulations in the public cloud. We perform meso-scale simulations of blood flow in image-reconstructed capillaries, and examine targeted drug delivery by artificial bacterial flagella (ABFs). The simulations deploy dissipative particle dynamics (DPD) with two software frameworks, Mirheo (developed by our team) and LAMMPS. Mirheo exhibits remarkable weak scalability for up to 512 GPUs. Similarly, LAMMPS demonstrated excellent weak scalability for pure solvent as well as for blood suspensions and ABFs in reconstructed retinal capillaries. In particular, LAMMPS maintained weak scaling above 90% on the cloud for up to 2,000 cores. Our findings demonstrate that cloud computing can support tightly coupled, large-scale scientific simulations with competitive performance.

Paper Structure

This paper contains 15 sections, 9 equations, 9 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Dissipative Particle Dynamics
  4. Red blood cell membranes
  5. Artificial Bacterial Flagella
  6. Implementation and Parallelization strategies
  7. Implementation of Red Blood Cell membranes
  8. Implementation of Artificial Bacterial Flagella
  9. Description of the compute environments
  10. Performance of Mirheo on the cloud
  11. Performance of LAMMPS on the cloud
  12. Cost efficiency
  13. ABF in a retinal capillary network
  14. Limitations
  15. Summary

Figures (9)

  • Figure 1: Weak scaling of Mirheo on the cloud with 4 T4 GPUs per node and on Piz Daint with one P100 GPU per node. : Piz daint, $(96r_c)^3$ volume per GPU. : GCP, $(96r_c)^3$ volume per GPU. : GCP, $(128r_c)^3$ volume per GPU. Top: DPD particles only. Bottom: Bulk blood.
  • Figure 2: Strong scaling of Mirheo on the cloud with 4 T4 GPUs per node. : $(256r_c)^3$ volume, : $(384r_c)^3$ volume, : ideal. Top: DPD particles only. Bottom: Bulk blood.
  • Figure 3: Weak scaling performance of LAMMPS. Top: DPD only, $(128r_c)^3$ volume per node, : GCP c2-standard-60 partition, 30 cores per node, : Piz Daint GPU-partition 12 cores per node, : Piz Daint multi-core partition, 36 cores per node. Bottom: Blood suspension with $30\%$ hematocrit, $(128r_c)^3$ volume per node, : GCP c2-standard-60 partition, 30 cores per node, : Piz Daint GPU-partition 12 cores per node.
  • Figure 4: Weak scaling of LAMMPS on the C2 partition: load imbalance for a volume of $(128r_c)^3$ per node. : Pure DPD particles, : Blood suspension with $30\%$ hematocrit.
  • Figure 5: Strong scaling of LAMMPS on the C2 partition with $(128r_c)^3$ volume. : DPD particles only. : Bulk blood at $30\%$ hematocrit. : ideal.
  • ...and 4 more figures