Taming My Noisy Homelab — Climacs IT Consulting

The Problems

Two Devices, Two Noise Sources

What seemed like hardware failure turned out to be workload behavior in both cases.

Storage Noise

UGREEN NAS

HDD Clicking & Chatter

Mechanical clicking from HDD-backed Docker storage
Gitea + Gitea Runner generating frequent small writes
Repeated runner retries creating constant I/O churn
Random writes on HDD are unexpectedly loud
Stopping containers silenced the disks entirely

Thermal Noise

Lenovo M70q Gen 5 · i5-14600T

Proxmox Fan Ramping

Fan spinning up aggressively every few seconds
Heavy background VMs spread across all cores
Board temp at 76°C, CPU package at 69°C
Pinning background VMs to specific CPU lanes reduced heat
Workload placement in SFF hosts matters enormously

Results

Before vs After

Measurable improvements from software-only changes — on both devices.

UGREEN NAS — Storage Noise

Before

Constant

HDD Clicking & Chatter

24/7

Gitea + Runner I/O Churn

Disks never resting

After

Silent

HDD Noise After Fix

Idle

Containers Stopped

Silent disks

Proxmox · Lenovo M70q — Thermal Noise

Before

76°C

Board / ACPI Temp

69°C

CPU Package Temp

Fan ramping every few seconds

After

56°C

Board / ACPI Temp

56°C

CPU Package Temp

Calm fan · Steady temps

Diagnosis

Root Cause Troubleshooting

Two parallel investigation paths that converged on the same lesson.

NAS — Storage Path

Hear clicking noise from NAS

Inspect disk I/O with iotop

Inspect Docker mounts via docker inspect

Inspect runner logs — retries/errors found

Stop Gitea + Gitea Runner containers

Clicking stops immediately

Root cause confirmed ✓

Proxmox — Thermal Path

Notice aggressive fan ramping

Check temps with sensors

Inspect load with htop

Identify heavy KVM process PIDs

Pin background VMs with taskset

Temps drop ~20°C

Fan behavior normalizes ✓

Tuning Models

Workload Placement Models

Practical models to separate priority workloads from background noise, and chatty I/O from mechanical disks.

NAS Storage Placement Model

High-IOPS PoolNVMe SSD

1TB NVMe

Docker AppData

Gitea Runners

Bulk Data PoolSATA HDD Array

16TB HDD

Media Library

Backups

Moving chatty databases and runner caches to SSDs prevents random reads/writes from keeping the mechanical HDDs awake.

Proxmox CPU Affinity Model

Priority LaneCPUs 0-11

0

1

2

3

4

5

6

7

8

9

10

11

Web VM

Database

Background LaneCPUs 12-19

12

13

14

15

16

17

18

19

K8s VMs

OpenRAG VM

Background workloads pinned to E-cores reduced thermal pressure on performance-critical lanes

Commands

Key Commands Used

A handful of targeted commands that revealed the root causes.

UGREEN NAS

terminal · nas

# === PHASE 1: DIAGNOSIS ===
# Watch cumulative disk I/O to find the chattiest processes
$ sudo iotop -oPa

# Query which files the suspicious process (e.g. PID 441) holds open
$ sudo lsof -p 441

# We isolated the noise to Gitea. Let's trace where its data lives physically.
$ docker inspect gitea | grep -A 20 "Mounts"



# === PHASE 2: RESOLUTION ===
# Temporarily stop the noisy containers
$ docker stop gitea gitea-runner

# Edit the application configuration to dial back the aggressive logging
$ sudo nano /volume1/@docker/volumes/.../conf/app.ini

# Restart the containers with the quieted configuration
$ docker start gitea gitea-runner
                        

Proxmox · i5-14600T

terminal · proxmox

# === PHASE 1: DIAGNOSIS ===
# Check raw hardware thermal sensors to verify CPU package temps
$ sensors

# Find the exact process ID (PID) of the heavy background Virtual Machine
$ ps aux | grep "/usr/bin/kvm -id 310"



# === PHASE 2: RESOLUTION ===
# Immediately (but temporarily) test the fix by pinning the live VM
$ taskset -cp 12-19 <PID>

# Make the CPU affinity permanent for that specific VM
$ qm set 310 --affinity 12-19

# Pin high-priority Database container (LXC) to performance cores
$ pct set 800 --cpuset 0-11
                        

Resolutions

The Fixes & Next Steps

Turning diagnosis into actionable improvements for a quieter lab.

NAS Fix Strategy

1. Immediate

Stop Gitea Runner manually when not actively running CI pipelines.

2. Tuning

Reduce noise in app.ini by setting LEVEL=warn and disabling router/SQL logs.

3. Ideal

Move chatty Docker app data off mechanical drives and onto SSD/NVMe.

Engineering Follow-Through

NAS Implement "quiet-hours" using cron routines to gracefully spin down the dev stack at night.
Proxmox Enforce CPU placement persistently at boot using documented systemd service overrides.

Hardware noise is often a software symptom.

Developer tools on HDD-backed storage can be surprisingly loud
Compact hosts react strongly to unfocused background load
Observability + smart placement = dramatically quieter homelab

The final lesson: Storage tiering, workload isolation, and diligent observability matter just as much in a small homelab as they do in enterprise environments. Treating your home infrastructure like a production system prevents "spooky behavior" and transforms your living room space back into the quiet environment it was meant to be.

Linux

Docker

Proxmox

iotop

sensors

taskset

htop