Spiderpool: High-Performance Underlay Networking for Kubernetes

Introduction

Most Kubernetes networking conversations are about overlays, service discovery, and “does it work?” Spiderpool lives in the other lane: when you care about squeezing latency and throughput out of the underlying network—especially for AI and HPC-style workloads.

Spiderpool, accepted as a CNCF Sandbox project in 2024, is a Kubernetes network solution offering underlay and Remote Direct Memory Access (RDMA) capabilities. With IPv4 and IPv6 support, it targets environments where high-performance networking and predictable data paths matter.

Underlay Networking

• Direct networking provides direct access to physical network infrastructure.
• Bypass overlay enables bypassing overlay network overhead.
• Performance optimization provides optimal network performance.
• Low latency minimizes network latency for high-performance workloads.

RDMA Support

1. RDMA capabilities enable Remote Direct Memory Access for high-performance networking.
2. Zero-copy networking enables zero-copy data transfers.
3. Low latency provides minimal latency for RDMA operations.
4. High throughput enables high throughput for data-intensive workloads.

IPv4 and IPv6 Support

• Dual-stack support enables both IPv4 and IPv6 networking.
• Migration support enables migration between IPv4 and IPv6.
• Compatibility ensures compatibility with existing network infrastructure.
• Flexibility provides flexibility in network configuration.

AI Workload Optimization

• High bandwidth enables high bandwidth for AI data transfers.
• Low latency minimizes latency for AI training workloads.
• Scalability enables scaling to large AI clusters.
• Performance monitoring provides visibility into network performance.

Use Cases

• AI training enables high-performance networking for AI training.
• HPC workloads enable high-performance computing workloads.
• Data intensive applications enable data-intensive applications.
• Real-time applications enable real-time applications requiring low latency.

Practical notes (underlay + RDMA comes with constraints)

• Hardware and drivers matter: RDMA isn’t a “toggle”—NIC capabilities, kernel/driver versions, and node configuration often define what’s feasible.
• IPAM and address planning: underlay approaches can put pressure on IP capacity and routing design; plan address management up front.
• Observability is non-negotiable: when performance is the goal, you’ll want clear signals for latency, drops, and saturation to avoid blind tuning.

Summary

Spiderpool represents a significant advancement in Kubernetes networking, providing teams with high-performance networking capabilities for demanding workloads.