SoarAlto

KO | EN

gitlite — search

#react #ai #typescript #llm #android #rust #nodejs #java #windows #awesome #pytorch #ios #openai #open-source

(★ 37)

Tiered Memory Management Beyond Hotness (OSDI'25)

SoarAlto Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Tiered Memory Management Beyond Hotness

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

This repository contains the implementation and evaluation artifacts for our
OSDI 2025 paper **"Tiered Memory Management Beyond Hotness"**. It presents two
tiered memory management policies: SOAR for memory allocation, and ALTO for
page migration..

## Repository Structure

### Core Components

- **[`src/alto/`](src/alto/)** - **ALTO** (AOL-based Layered Tiering Orchestration) implementation
  - Kernel patches for various tiering systems with ALTO integration
  - Supports TPP, NBT, Nomad, and Colloid tiering systems

- **[`src/soar/`](src/soar/)** - **SOAR** (Static Object Allocation based on Ranking) implementation
  - Profiling and analysis tools for object-level memory placement
  - Allocation interception and control mechanisms

- **[`src/microbenchmark/`](src/microbenchmark/)** - Synthetic workloads for motivation
  - Pointer-chasing and sequential memory access patterns

### Runtime and Evaluation

- **[`run/`](run/)** - Experiment orchestration scripts
  - Automated setup and configuration
  - Benchmark execution for supported systems
  - Performance monitoring and data collection

## Supported Systems

Our implementation supports the following tiered memory management systems:

| System | Description | ALTO Support |
|--------|-------------|--------------|
| **TPP** | Transparent Page Placement, ASPLOS'23 | ✅ |
| **NBT** | Linux NUMA-Balancing-Tiering | ✅ |
| **Nomad** | Non-exclusive Memory Tiering, OSDI'24 | ✅ |
| **Colloid** | Access Latency is Key!, SOSP'24 | ✅ |

## Testing Platforms

Our evaluation was conducted on the following platforms:

- **SKX**: Two Intel Xeon Silver 4114 10-core CPUs at 2.20 GHz, 192GB DDR4 Memory
- **SPR**: Two Intel Xeon Gold 6430 32-core CPUs at 2.10 GHz, 256GB DDR5 Memory

## Installation and Setup

### Quick Setup

```bash
cd run
./setup.sh
```

This script will:
- Configure system settings (disable THP, NUMA balancing, etc.)
- Build necessary kernel modules
- Set up monitoring tools

### Detailed Configuration

For detailed setup instructions for each component, see:
- [ALTO Setup Guide](src/alto/README.md)
- [SOAR Setup Guide](src/soar/README.md)
- [Runtime Configuration](run/README.md)

## Running Experiments

### Basic Usage

```bash
cd run
./run.sh [type] [threads] [MLC-threads-list]
```

**Parameters:**
- `type`: System type (0-11, see table below)
- `threads`: Number of application threads
- `MLC-threads-list`: Comma-separated list of MLC thread counts

**System Types:**
```
0: NoTier        5: TPP-ALTO      10: Remote
1: TPP           6: NBT-ALTO      11: SOAR
2: NBT           7: Nomad-ALTO
3: Nomad         8: Colloid-ALTO
4: Colloid       9: Local
```

### Example Experiments

```bash
# Run TPP-ALTO with 4 application threads
./run.sh 5 4 0,1,2

# Compare baseline vs ALTO-enhanced systems
./run.sh 1 4 0,1,2  # TPP baseline
./run.sh 5 4 0,1,2  # TPP with ALTO

# SOAR object-level allocation
./run.sh 11 4 0,1,2
```

## Performance Analysis

### Data Collection

The scripts automatically collect:
- Page promotion statistics
- System performance metrics (via `perf`)
- Application-specific metrics

### Analysis Tools

- `calpg.sh`: Calculate page promotion counts
- `proc_obj_e.py`: Analyze object allocation patterns (for SOAR)
- Performance visualization scripts (in respective component directories)

## Research Components

### ALTO (AOL-based Layered Tiering Orchestration)

ALTO provides a unified framework for regulating page promotion across different tiering systems. Key features:

### SOAR (Static Object Allocation based on Ranking)

SOAR enables object-level memory placement decisions based on access pattern analysis:

- **Profiling phase**: Tracks object access patterns
- **Ranking algorithm**: Prioritizes objects for local/remote placement
- **Runtime placement**: Intercepts allocation calls for optimal placement

## Citation

If you use this work in your research, please cite our OSDI 2025 paper:

```bibtex
@inproceedings{SoarAlto.osdi25,
  author       = {Jinshu Liu and Hamid Hadian and Hanchen Xu and Huaicheng Li},
  title        = {Tiered Memory Management Beyond Hotness},
  booktitle    = {In the 19th USENIX Symposium on Operating Systems Design and Implementation, {OSDI} 2025, Boston, MA, USA, July 7-9, 2025},
  pages        = {731--747},
  publisher    = {{USENIX} Association},
  year         = {2025},
  url          = {https://www.usenix.org/conference/osdi25/presentation/liu},
}
```

## Contact

**Maintainer**: Jinshu Liu - [jinshu@vt.edu](mailto:jinshu@vt.edu)

For questions about the research or implementation details, please open an
issue or contact the maintainer directly.

## License

This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.

SoarAlto

SoarAlto Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

SoarAlto

SoarAlto Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README