Skip to content

About RHAPSODY

RHAPSODY (Runtime for Heterogeneous APplications, Service Orchestration and DYnamism) is a cutting-edge runtime system designed to address the evolving challenges of modern scientific computing. As computational workflows become increasingly complex, combining traditional HPC simulations with AI/ML workloads, there is a critical need for systems that can efficiently orchestrate heterogeneous tasks across diverse computing resources.

Vision

Our vision is to create a unified runtime that seamlessly bridges the gap between traditional high-performance computing and emerging AI/ML paradigms, enabling scientists to focus on their research rather than the complexities of workflow orchestration.

Core Philosophy

RHAPSODY is built on several key principles:

Heterogeneity as a First-Class Citizen

Modern scientific workflows are inherently heterogeneous, combining:

  • Traditional HPC simulations (CPU-intensive)
  • Machine learning training and inference (GPU-intensive)
  • Data processing and analysis tasks
  • I/O-intensive operations

RHAPSODY treats this heterogeneity as a fundamental characteristic rather than an afterthought.

Dynamic Adaptation

Scientific workflows are rarely static. RHAPSODY supports:

  • Runtime modification of task graphs based on intermediate results
  • Adaptive resource allocation
  • Dynamic load balancing across heterogeneous resources

Platform Abstraction

Scientists should not need to be experts in every computing platform. RHAPSODY provides:

  • Unified interfaces across different HPC systems
  • Abstraction of resource manager specifics
  • Portable workflow descriptions

Technical Innovation

RHAPSODY introduces several technical innovations:

Pluggable Backend Architecture

The system supports multiple execution backends, each optimized for different scenarios:

  • Concurrent Backend: For shared-memory parallel execution
  • Dask Backend: For distributed Python workloads
  • RADICAL-Pilot Backend: For large-scale HPC execution

AsyncFlow Integration

Full compatibility with the AsyncFlow workflow management system, providing:

  • Standardized workflow descriptions
  • Tool interoperability
  • Community ecosystem benefits

State-of-the-Art Monitoring

Comprehensive monitoring and introspection capabilities for:

  • Real-time task execution tracking
  • Performance analysis and optimization
  • Fault detection and recovery

Research Impact

RHAPSODY enables breakthrough research in numerous domains:

  • Climate Science: Coupling atmospheric simulations with ML-based post-processing
  • Materials Science: Integrating quantum simulations with ML property prediction
  • Computational Biology: Combining molecular dynamics with deep learning analysis

Project Timeline

RHAPSODY development follows a structured timeline aligned with NSF project milestones:

  • Phase 1 (2021-2022): Core architecture and backend development
  • Phase 2 (2022-2024): Integration with existing workflow systems
  • Phase 3 (2024-2026): Production deployment and optimization

Team and Collaboration

RHAPSODY is developed by the RADICAL Research Team at Rutgers University, in collaboration with:

  • DOE National laboratories
  • Campus supercomputing centers
  • Academic research institutions

Open Science Commitment

We are committed to open science principles:

  • Open Source: All code is publicly available under MIT license
  • Open Data: Benchmarks and performance data are shared publicly
  • Open Standards: We contribute to community standards and protocols
  • Open Collaboration: We welcome contributions from the broader community

Future Directions

Looking ahead, RHAPSODY will continue to evolve to meet emerging needs:

  • Enhanced support for quantum computing integration
  • Advanced AI/ML workflow optimization
  • Edge computing and IoT integration for scientific workflows
  • Expanded ecosystem of compatible tools and frameworks