⚙️ Infrastructure Advanced ⏱️ 15 min

Homelab Server & NAS Setup with TrueNAS Scale

A comprehensive guide to the hardware powering BlueRobin. From Ryzen processors and 128GB RAM to configuring TrueNAS Scale for Kubernetes storage via iSCSI and NFS.

By Victor Robin •
infrastructurehardwarestorage

Introduction

High-performance software needs high-performance hardware. When we decided to bring BlueRobin on-premise, we needed a foundation that could handle the simultaneous demands of Kubernetes virtualization, massive storage throughput for MinIO, and spare cycles for build agents. We chose a consolidated approach using a powerful workstation build running TrueNAS Scale.

Why Proper Hardware Matters:

  • Reliability: ECC memory and ZFS protect against bit rot.
  • Bandwidth: 10GbE overcomes the “network bottleneck,” making remote storage feel local.
  • Density: Running everything on one optimized rig saves power vs. a rack of noisy 1U servers.

What We’ll Build

In this guide, we will walk through the specifications and configuration of the BlueRobin homelab server. You will learn how to:

  1. Select Hardware: Balancing CPU cores vs. PCIe lanes.
  2. Configure ZFS: Setting up pools for speed (NVMe) and capacity (HDD).
  3. Expose Storage: Connecting TrueNAS to Kubernetes via NFS and iSCSI.

Architecture Overview

The homelab infrastructure connects enterprise-grade hardware with a Kubernetes cluster for seamless storage provisioning:

flowchart TB
    subgraph Hardware["🖥️ Server Hardware"]
        CPU["⚡ Ryzen 9 5950X\n16C/32T"]
        RAM["🧠 128GB DDR4 ECC"]
        NIC["🔌 Intel X520-DA2\n10GbE SFP+"]
    end

    subgraph ZFS["💾 ZFS Pools"]
        Fast["⚡ tank-fast\nNVMe Mirror\nVMs & Databases"]
        Bulk["📦 tank-bulk\nHDD RaidZ1\nArchives & Backups"]
    end

    subgraph Network["🌐 Network (10GbE)"]
        DAC["🔗 DAC Cable\n9.4 Gbps"]
        Switch["📡 10GbE Switch"]
    end

    subgraph K3s["☸️ K3s Cluster"]
        CSI["🔧 Democratic-CSI"]
        NFS["📂 NFS Volumes\n(ReadWriteMany)"]
        iSCSI["💽 iSCSI Volumes\n(ReadWriteOnce)"]
    end

    CPU --> RAM
    RAM --> ZFS
    Fast --> NIC
    Bulk --> NIC
    NIC --> DAC
    DAC --> Switch
    Switch --> CSI
    CSI --> NFS
    CSI --> iSCSI

    classDef primary fill:#7c3aed,color:#fff
    classDef secondary fill:#06b6d4,color:#fff
    classDef db fill:#f43f5e,color:#fff
    classDef warning fill:#fbbf24,color:#000

    class Hardware primary
    class K3s secondary
    class ZFS db
    class Network warning

Data Flow:

  1. Hardware Layer: High-core-count CPU and ECC RAM provide reliability
  2. Storage Layer: ZFS pools tiered by performance (NVMe) and capacity (HDD)
  3. Network Layer: 10GbE eliminates the storage bottleneck
  4. Kubernetes Layer: Democratic-CSI dynamically provisions volumes via NFS/iSCSI

Hardware Architecture

Our build centers on the AMD Ryzen platform, utilizing its high core count and ECC memory support (unofficial but functional on AsRock Rack boards).

graph TD
    subgraph Server["Home Server (TrueNAS Scale)"]
        CPU[Ryzen 9 5950X<br/>16 Cores / 32 Threads]
        RAM[128GB DDR4 ECC UDIMM]
        HBA[LSI 9300-8i HBA]
        NIC[Intel X520-DA2<br/>Dual 10GbE SFP+]
        
        subgraph Storage["ZFS Pools"]
            PoolSpeed[NVMe Pool<br/>2x 2TB Gen4 Mirror<br/>VMs & Databases]
            PoolBulk[HDD Pool<br/>4x 14TB RaidZ1<br/>Archives & Backups]
        end
        
        CPU --- RAM
        CPU --- HBA
        HBA --- PoolBulk
        CPU --- PoolSpeed
        CPU --- NIC
    end

    classDef primary fill:#7c3aed,color:#fff
    classDef secondary fill:#06b6d4,color:#fff
    classDef db fill:#f43f5e,color:#fff
    classDef warning fill:#fbbf24,color:#000

    class CPU,RAM,HBA,NIC primary
    class Server secondary
    class PoolSpeed,PoolBulk db

Section 1: The Hardware Selection

The Brain: CPU & RAM

We chose the Ryzen 9 5950X for its incredible multi-threaded performance. Kubernetes creates many small containers, and having 32 threads allows us to over-provision efficiently.

128GB of RAM is critical. ZFS loves RAM for its Adaptive Replacement Cache (ARC), and running a K3s cluster, Postgres, Qdrant, and local AI models eats memory for breakfast.

The Spine: Networking

10GbE is non-negotiable. We use standardized Intel X520 cards with DAC (Direct Attach Copper) cables. This provides a steady 9.4 Gbps link between the server and our main workstation, enabling us to edit video directly off the NAS or restore gigantic database backups in seconds.

Section 2: TrueNAS Scale Configuration

TrueNAS Scale (Linux-based) was preferred over Core (BSD-based) for better Docker/Kubernetes compatibility and KVM support.

Storage Tiers

We split our storage into two distinct pools:

  1. tank-fast (NVMe Mirror):
    • Use Case: Docker volumes, Postgres DB files, Qdrant vectors.
    • Why: IOPS. Random I/O on HDDs kills database performance.
  2. tank-bulk (HDD RaidZ1):
    • Use Case: MinIO objects, Backups, Media.
    • Why: Cost-effective bulk storage.

Section 3: Connecting to Kubernetes

We use Democratic-CSI (configured via Flux) to automatically provision storage on TrueNAS from our K3s cluster.

NFS vs iSCSI

  • NFS: Used for ReadWriteMany volumes (e.g., shared configs, web assets). Faster to set up, easier to debug.
  • iSCSI: Used for ReadWriteOnce volumes (e.g., Databases). Block-level access provides better performance and consistency guarantees for PostgreSQL.
# Example StorageClass for TrueNAS iSCSI
apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: truenas-iscsi
provisioner: org.democratic-csi.iscsi
parameters:
  fsType: ext4
  # ... configuration linking to TrueNAS API

Conclusion

This hardware setup provides the bedrock for the entire BlueRobin platform. By investing in quality components like ECC RAM and 10GbE, we’ve built a “private cloud” that outperforms standard cloud instances at a fraction of the long-term cost.

Next Steps: