root
91 lines
2.5 KiB
Markdown
91 lines
2.5 KiB
Markdown
# pve2 — Host-Infrastruktur
|
||
|
||
**IP:** 192.168.10.4 · **GPU:** 2× NVIDIA GeForce GTX 1080 · **Treiber:** 580.95.05
|
||
|
||
## Physische Disks
|
||
|
||
| Device | Größe | Nutzung |
|
||
|--------|-------|---------|
|
||
| nvme0n1 | ~477 GB | System, `local-lvm` |
|
||
| nvme1n1 | ~477 GB | `nvme_second` (z. B. AIDEV) |
|
||
| sda | 1,8 TB | `records` — Aufnahmen, Backups, Docker-Daten |
|
||
|
||
Details: [01_System-und-Speicher-Uebersicht.md](01_System-und-Speicher-Uebersicht.md)
|
||
|
||
## Storage-Pools
|
||
|
||
| Pool | Typ | Inhalt |
|
||
|------|-----|--------|
|
||
| local | dir | ISO, Templates |
|
||
| local-lvm | lvmthin | VM/CT auf nvme0 |
|
||
| nvme_second | lvmthin | VM/CT auf nvme1 |
|
||
| records | dir | HDD — Backups, Frigate, große Daten |
|
||
|
||
## VMs (Auswahl)
|
||
|
||
| VMID | Name | Rolle |
|
||
|------|------|-------|
|
||
| 104 | opnsense | **Router/Firewall** — produktiv |
|
||
| 106 | homeassistant | Home Assistant + Mosquitto |
|
||
|
||
## Container (Auswahl)
|
||
|
||
| CTID | Name | GPU | Rolle |
|
||
|------|------|-----|-------|
|
||
| 101 | docker | **Nein** (NVIDIA entfernt) | Frigate, Compose-Stack |
|
||
| 109 | media | optional | Medien (oft gestoppt) |
|
||
| 110 | AIDEV | **Ja** | Jupyter/ML |
|
||
|
||
### GPU-Mount-Policy LXC
|
||
|
||
| CT | `/dev/nvidia*` | Grund |
|
||
|----|----------------|-------|
|
||
| 101 | **Nein** | Frigate = OpenVINO + Intel VAAPI |
|
||
| 110 | **Ja** | ML bei Bedarf |
|
||
| 109 | nur wenn aktiv | Gestoppt → kein Mount |
|
||
|
||
Konfiguration: `/etc/pve/lxc/101.conf` — NVIDIA-Zeilen auskommentiert (`#lxc.mount.entry%3A ...`).
|
||
|
||
Frigate in CT 101:
|
||
- Detector: OpenVINO (CPU/iGPU)
|
||
- `hwaccel_args: preset-vaapi`
|
||
- Devices: `/dev/dri/renderD128`, `/dev/dri/card0`
|
||
- NVIDIA in `compose.yml` auskommentiert
|
||
|
||
## NVIDIA auf dem Host
|
||
|
||
```bash
|
||
nvidia-smi
|
||
systemctl status nvidia-persistenced
|
||
```
|
||
|
||
Persistence Mode **Pflicht** für P8 Idle (~8 W/GPU). Service-Dateien auch im Repo server-power: `deploy/nvidia-persistenced/`.
|
||
|
||
Vollständige GPU-Doku: [09_GPU-Idle-vollstaendig.md](09_GPU-Idle-vollstaendig.md)
|
||
|
||
## Host-Dienste
|
||
|
||
| Dienst | Zweck |
|
||
|--------|-------|
|
||
| `nvidia-persistenced` | GPU Persistence Mode |
|
||
| `pve-power-mqtt` | RAPL + nvidia-smi → MQTT |
|
||
| Proxmox | Web :8006 |
|
||
|
||
## Git / Doku auf diesem Host
|
||
|
||
| Pfad | Inhalt |
|
||
|------|--------|
|
||
| `/root/docu` | docu-Repo (Git) |
|
||
| `/root/code/pve-power-mqtt` | Go-Agent + GPU-Doku |
|
||
| `/root/.git-credentials-jeanavril` | Gitea Token |
|
||
|
||
## Ansible
|
||
|
||
Playbooks: siehe [02_Ansible-Playbooks.md](02_Ansible-Playbooks.md)
|
||
|
||
## Wartung
|
||
|
||
- `fstrim` in VMs/CTs für Thin-Pools
|
||
- Backup auf `records`, nicht `local-lvm`
|
||
- Speicher-Monitoring: [05_Wartung-und-Monitoring.md](05_Wartung-und-Monitoring.md)
|