Kube (Kubernetes) and dAI( Data and AI)
Welcome to Kube-dAI, an open-source project dedicated to providing highly scalable deployment patterns and infrastructure-as-code templates for data processing and AI workloads on Kubernetes.
Kube-dAI aims to empower organizations with battle-tested, scalable architectures for running data and AI workloads on Kubernetes. Our project offers:
- Highly scalable deployment patterns
- Best practices for compute and storage optimization
- Comprehensive benchmarks for performance evaluation
- Infrastructure-as-code templates for rapid deployment
Whether you're orchestrating large-scale data processing pipelines or deploying cutting-edge AI models, Kube-dAI provides the insights and tools you need to maximize your Kubernetes infrastructure.
Our project delves deep into compute and storage best practices for each deployment, whether it's data processing or generative AI inference. We cover two main categories:
- Apache Spark
- Apache Flink
- Trino (formerly PrestoSQL)
- KServe
- Ray
- NVIDIA Triton Inference Server
- LWS (Leader Worker Set)
For each technology, we provide:
- Scalable architecture designs
- Performance optimization techniques
- Resource allocation strategies
- Storage configuration best practices
- Comprehensive benchmarks
Each repository includes detailed benchmarks and best practices, covering:
- Scalability tests (from gigabytes to petabytes)
- Throughput and latency measurements
- Resource utilization optimizations
- Cost-efficiency analyses
- Comparative studies of storage solutions
- Best practices for specific workload types
Our GitHub organization hosts multiple repositories, each dedicated to a specific technology:
Data Processing:
spark-on-k8s
: Apache Spark deployment patterns and benchmarksflink-on-k8s
: Apache Flink deployment patterns and benchmarkstrino-on-k8s
: Trino deployment patterns and benchmarks
Artificial Intelligence:
kserve-on-k8s
: KServe deployment patterns and benchmarksray-on-k8s
: Ray deployment patterns and benchmarkstriton-on-k8s
: NVIDIA Triton Inference Server deployment patterns and benchmarkslws-on-k8s
: LWS deployment patterns and benchmarks
Additional Repositories:
genai-inference-patterns
: Generative AI inference optimization techniquesdata-storage-benchmarks
: Comparative studies of storage solutions for big dataai-model-serving-best-practices
: Best practices for serving AI models at scale
To leverage Kube-dAI for your projects:
- Explore our repositories to find relevant technologies and patterns.
- Clone the repository of interest:
git clone https://github.com/Kube-dAI/<repository-name>.git
- Follow the detailed README in each repository for setup, benchmarking, and best practices.
We welcome contributions from the community! To contribute:
- Fork the relevant repository.
- Create a new branch for your feature, benchmark, or optimization.
- Submit a pull request with your improvements.
Please review our Contribution Guidelines for more details.
This project is open-source and available under the Apache License 2.0.
For assistance or to discuss advanced use cases:
- Open an issue in the relevant repository
- Join our community Slack (placeholder link)
- Explore our comprehensive documentation (placeholder link)
- Project Website (placeholder link)
- Documentation (placeholder link)
- Community Forum (placeholder link)
- Blog (placeholder link) - For the latest benchmarks and best practices
Kube-dAI: Empowering Scalable Data and AI Deployments on Kubernetes