Lab Philosophy

This lab is designed around clarity, longevity, and predictability.

Systems are built to run continuously for years with minimal intervention, favoring conservative architecture decisions over maximum density or short-term performance gains.

Guiding Principles

Single-Socket by Design

All systems use single-socket platforms.

This avoids:

• NUMA complexity
• cross-socket latency
• unnecessary scheduling overhead

Modern single-socket EPYC systems provide more than enough memory bandwidth and PCIe connectivity for the workloads in this environment.

Predictable Over Peak

Peak benchmarks are less interesting than behavior under sustained load.

Hardware is evaluated based on:

• thermal stability
• I/O consistency
• ease of troubleshooting
• long-term supportability

Mirrors Over Complexity

ZFS mirrors are preferred over complex parity layouts.

This simplifies:

• resilvering
• expansion planning
• failure isolation

The goal is not maximum usable capacity, but operational clarity.

Hardware as Infrastructure

Systems are treated as infrastructure, not experiments.

Changes are intentional, documented, and incremental.

If a component cannot justify its complexity over time, it does not belong in the lab.

Closing Notes

The systems documented here are not designed to be impressive on paper.

They are designed to be boring, reliable, and understandable — even years after deployment.