README.md

SysPerf: A Supercomputing Benchmark Framework

SysPerf is a passion project for benchmarking designed to provide comprehensive performance analysis for HPC environments. The current implementation focuses on storage performance testing via FIO, with a clear path toward complete cluster performance evaluation capabilities.

Current Features

Storage System Analysis

• FIO Integration
  • Block device performance testing
  • Configurable I/O patterns
  • Multi-threaded I/O testing
  • IOPS and throughput measurements
  • Latency analysis
  • Queue depth impact assessment

On Deck Features

• Results Processing
  • Real-time metric collection
  • Embedded database storage
  • Basic visualization capabilities
  • JSON/CSV export options

Quick Start

Prerequisites

• Rust 1.70 or higher
• FIO 3.x or higher
• Linux kernel 4.x or higher

Installation

# Clone the repository
git clone https://github.com/kennethdsheridan/sysperf-svr
cd sysperf-svr

# Build the project
cargo build --release

# Run tests
cargo test

Basic Usage

# Run basic FIO test
./target/release/sysperf-svr benchmark --tool fio &

Configuration

Example configuration for storage testing:

# config.yaml
storage_benchmark:
  fio:
    block_sizes: 
      - "4k"
      - "8k"
      - "64k"
      - "1m"
    io_engine: "libaio"
    runtime: 60
    iodepth: [1, 8, 16, 32, 64]
    numjobs: [1, 4, 8]
    test_types:
      - "random-read"
      - "random-write"
      - "sequential-read"
      - "sequential-write"

database:
  embedded:
    path: "./data"
    retention_days: 30
  remote:
    enabled: false
    type: "influxdb"
    url: "http://metrics-db:8086"

Architecture

SysPerf follows a Hexagonal (Ports & Adapters) design. At a glance, the project is organised like this:

sysperf-svr/
├── src/
│   ├── adapters/                 # ⬅️ OUTBOUND ADAPTERS (secondary)
│   │   ├── benchmark_adapter.rs
│   │   ├── database_adapter.rs
│   │   ├── log_adapter.rs
│   │   ├── metrics_adapter.rs
│   │   └── mod.rs
│   │
│   ├── application/              # ⬅️ APPLICATION SERVICES (use-cases)
│   │   ├── metrics/              #  ├─ orchestrates metric-collection use-cases
│   │   ├── storage/              #  ├─ orchestrates storage-benchmark flows
│   │   └── mod.rs
│   │
│   ├── cli/                      # ⬅️ PRIMARY ADAPTER – command-line interface
│   │   └── …                     #     (struct-opt/clap handlers live here)
│   │
│   ├── database/                 # ⬅️ DB bootstrap & migration helpers
│   │   └── …                     
│   │
│   ├── domain/                   # ⬅️ ENTERPRISE DOMAIN (pure business rules)
│   │   └── …                     
│   │
│   ├── ports/                    # ⬅️ PORTS (traits) – the “hexagon edges”
│   │   ├── benchmark_port.rs
│   │   ├── database_port.rs
│   │   ├── log_port.rs
│   │   ├── metrics_port.rs
│   │   ├── storage_port.rs
│   │   └── mod.rs
│   │
│   ├── wasm/                     # ⬅️ PRIMARY ADAPTER – WebAssembly front-end
│   │   └── …                     
│   │
│   ├── lib.rs                    # Library entry (re-exports of core crates)
│   └── main.rs                   # Binary entry (bridges CLI ↔ application layer)
│
├── config/                       # YAML & TOML examples, baseline configs
└── tests/                        # Black-box & integration tests

Architectural Approach

The project uses a Ports and Adapters (Hexagonal) architecture that provides:

• Ports Layer: Trait-based interfaces that abstract core functionality

  • Benchmarking ports
  • Metric collection ports
  • Database interaction ports
  • Node communication ports
  • Hardware management ports

• Adapters Layer: Concrete implementations that are hot-swappable

  • FIO benchmarking adapters
  • Hardware metric collectors
  • SurrealDB persistence adapter
  • MAAS integration adapters
  • Logging implementations

Key Benefits

• Component independence and hot-swappability
• Simplified testing through trait mocking
• High availability through adapter redundancy
• Easy extension of functionality
• Clear separation of concerns

Current Implementation Status

• Stable

  • Storage benchmarking ports and adapters (FIO)
  • Basic metric collection
  • Database integration

• In Development

  • Network testing ports
  • GPU benchmarking interfaces
  • Additional storage adapters

Roadmap

Network Performance

• InfiniBand Testing Suite

  • RDMA performance analysis
  • Subnet Manager performance
  • QoS validation
  • Network congestion testing

• Network Tools Integration

  • iperf3 TCP/UDP testing
  • perftest RDMA metrics
  • OpenMPI benchmark suite
  • Custom network test framework

GPU Cluster Testing

• Containerized GPU Benchmarks

  • HPL-AI implementation
  • NVIDIA NCCL tests
  • GPU-Direct RDMA
  • Multi-node GPU communication

• ML/AI Performance Suite

  • Distributed training metrics
  • Model inference benchmarks
  • Memory bandwidth testing
  • Scaling efficiency analysis

Advanced Storage

• Parallel Filesystem Testing

  • Lustre performance suite
  • GPFS throughput analysis
  • BeeGFS benchmarking
  • Metadata performance testing

• Extended I/O Testing

  • IOR integration
  • mdtest implementation
  • Custom I/O patterns
  • Multi-client testing

System Integration

• Job Scheduler Integration
  • SLURM support
  • LSF compatibility
  • PBS integration
  • Resource allocation analysis

Advanced Analytics

• Machine Learning Integration
  • Performance prediction
  • Anomaly detection
  • Resource optimization
  • Trend analysis

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• The HPC community
• The Rust community

Citation

@software{sysperf2024,
  author = {Kenny (Knight) Sheridan},
  title = {SysPerf: A Supercomputing Benchmark Framework},
  year = {2024},
  publisher = {GitHub},
  url = {https://github.com/kennethdsheridan/sysperf-svr}
}