Add Presto GPU landing page README

presto/README.md

@@ -0,0 +1,195 @@
+# Presto GPU
+
+Welcome to the Presto GPU testing and benchmarking infrastructure! This directory contains everything needed to build, deploy, and test Presto with GPU-accelerated native execution using NVIDIA cuDF and Velox.
+
+## Overview
+
+This infrastructure enables running Presto with GPU-accelerated query execution, leveraging the GPU-native Velox execution engine powered by NVIDIA cuDF. For more details about the technology behind GPU acceleration, see the [NVIDIA Developer Blog: Accelerating Large-Scale Data Analytics with GPU-Native Velox and NVIDIA cuDF](https://developer.nvidia.com/blog/accelerating-large-scale-data-analytics-with-gpu-native-velox-and-nvidia-cudf/).
+
+For general information about this repository, see the [main README](https://github.com/rapidsai/velox-testing/blob/main/README.md).
+
+## Repository Structure and Version Compatibility
+
+Ensure you have the following directory structure:
+
+```
+├─ base_directory/
+  ├─ velox-testing           # https://github.com/rapidsai/velox-testing
+  ├─ presto                  # https://github.com/prestodb/presto
+  ├─ velox                   # https://github.com/facebookincubator/velox
+```
+
+All three repositories must be checked out as sibling directories. **Important:** Please use compatible commit hashes/branches across these repositories. There is no formal public matrix yet—refer to release notes, this README, and instructions in the dependent repositories for guidance.
+
+## Current Testing Status
+
+**TPC-H** is the primary benchmark suite currently used for testing Presto GPU functionality. The infrastructure supports comprehensive testing including:
+- Functional correctness testing
+- Performance benchmarking
+- CPU vs GPU comparison
+- Integration testing with Hive metastore
+
+**TPC-DS** support is also available and under active development.
+
+## Quick Start
+
+### Prerequisites
+
+- Sibling repo checkout (see structure above)
+- Docker and docker-compose installed
+- NVIDIA GPU, drivers, and compatible CUDA toolkit
+
+### Building and Starting Presto GPU
+
+1. Navigate to the scripts directory:
+   ```bash
+   cd velox-testing/presto/scripts
+   ```
+
+2. Set up your data directory (optional but recommended):
+   ```bash
+   export PRESTO_DATA_DIR=/path/to/your/benchmark/data
+   ```
+   > **Tip:** Add this export to your `~/.bashrc` to avoid setting it each time.
+
+3. Build dependencies (first time only):
+   ```bash
+   ./build_centos_deps_image.sh
+   ```
+   > **Note:** Only internal team members with credentials can fetch a pre-built image (`./fetch_centos_deps_image.sh`). For most users, building locally is required.
+
+4. Start Presto with GPU workers:
+   ```bash
+   ./start_native_gpu_presto.sh
+   ```
+
+5. Access the Presto web UI at http://localhost:8080
+
+### Running Tests
+
+Run integration test suite:
+```bash
+cd velox-testing/presto/scripts
+./run_integ_test.sh --help              # See all options
+./run_integ_test.sh --benchmark-type tpch
+```
+
+Or directly via pytest:
+```bash
+cd velox-testing/presto/testing/integration_tests
+pytest tpch_test.py
+```
+
+### Running Benchmarks
+
+1. Ensure the Presto GPU instance is running.
+
+2. Set up benchmark tables if needed:
+   ```bash
+   cd velox-testing/presto/scripts
+   ./setup_benchmark_data_and_tables.sh --help  # See all options
+   ```
+
+3. **Important:** Run ANALYZE TABLES on CPU Presto first:
+   ```bash
+   ./analyze_tables.sh
+   ```
+   > This step is necessary because aggregation for statistics collection is not yet supported on GPU. Collecting statistics improves performance and reduces OOM errors for GPU execution.
+
+4. Run benchmarks:
+   ```bash
+   ./run_benchmark.sh --help
+   ./run_benchmark.sh --benchmark tpch --schema-name <your_schema>
+   ```
+   > **Note:** The `--schema-name` flag is required to specify your target DB schema for the benchmark.
+
+## Directory Structure
+
+- **`docker/`** - Docker Compose configurations and Dockerfiles for different Presto variants
+- **`pbench/`** - Performance benchmarking utilities and TPC-H query definitions
+- **`scripts/`** - Shell scripts for building, deploying, and testing Presto
+- **`testing/`** - Python-based test framework using pytest
+  - `integration_tests/` - Functional correctness tests
+  - `performance_benchmarks/` - Performance testing infrastructure
+  - `common/` - Shared test utilities and query definitions
+
+## Available Presto Variants
+
+1. **Java Presto** - Pure Java coordinator and workers
+   ```bash
+   ./start_java_presto.sh
+   ```
+2. **Native CPU Presto** - Java coordinator with native CPU workers
+   ```bash
+   ./start_native_cpu_presto.sh
+   ```
+3. **Native GPU Presto** - Java coordinator with GPU-accelerated native workers
+   ```bash
+   ./start_native_gpu_presto.sh
+   ```
+
+## Benchmark Data
+
+The infrastructure uses Hive-style directory layouts with Parquet files. For TPC-H, the expected structure is:
+
+```
+benchmark_data/
+└─ tpch/
+   ├─ customer/
+   ├─ lineitem/
+   ├─ nation/
+   ├─ orders/
+   ├─ part/
+   ├─ partsupp/
+   ├─ region/
+   └─ supplier/
+```
+
+Generate new data or set up tables on existing data using the provided scripts.
+
+## Testing Different Scale Factors
+
+There are two common approaches:
+
+1. **Generate new data for your preferred scale factor:**
+   ```bash
+   cd velox-testing/presto/testing/integration_tests/scripts
+   ./generate_test_files.sh --scale-factor 100
+   ```
+   Then run the tests as shown above.
+
+2. **Switch between pre-generated scale factors with schemas:**
+   Register tables for any generated data at any scale factor as a new schema:
+   ```bash
+   cd velox-testing/presto/scripts
+   ./setup_benchmark_data_and_tables.sh --scale-factor 100 --schema-name tpch_sf100
+   ./run_integ_test.sh --benchmark-type tpch --schema-name tpch_sf100
+   ```
+   This allows you to run benchmarks and tests without regenerating data for each scale factor—simply specify the schema name.
+
+## Configuration
+
+Configuration files for Presto are managed through a template system in:
+- `docker/config/template/` - Configuration templates
+- `docker/config/params.json` - Parameters for template filling
+
+Configuration is generated automatically when starting Presto or running tests. Manual execution of `generate_presto_config.sh` is rarely needed:
+```bash
+cd velox-testing/presto/scripts
+./generate_presto_config.sh   # Only if making custom template changes
+```
+
+## Resources
+
+- [Main Repository README](https://github.com/rapidsai/velox-testing/blob/main/README.md)
+- [NVIDIA Blog: GPU-Native Velox and cuDF](https://developer.nvidia.com/blog/accelerating-large-scale-data-analytics-with-gpu-native-velox-and-nvidia-cudf/)
+- [Presto Official Documentation](https://prestodb.io/)
+- [Velox Documentation](https://facebookincubator.github.io/velox/)
+
+## Contributing
+
+See the [CONTRIBUTING.md](https://github.com/rapidsai/velox-testing/blob/main/CONTRIBUTING.md) in the main repository for contribution guidelines.
+
+## License
+
+See [LICENSE](https://github.com/rapidsai/velox-testing/blob/main/LICENSE) in the main repository.