Track 02 / Hardware Bundle Spec
3 Managers · 15 Workers
18-Node Cluster
Node hardware
01
Manager · ×3
4C/4T Mini PC32GB RAM / 1TB SSD
Cores4
Threads4
RAM32 GB
SSD1 TB
Power draw~28 W
6× LAN ports — 1 for WAN uplink, 5 dedicated one-to-one to each worker in its bundle.
View listing — N150 / 32GB / 1TB →
Worker · ×15
Ryzen 7 7730U16GB RAM / 1TB SSD
Cores8
Threads16
RAM16 GB
SSD1 TB
Power draw~15 W
ACEMAGIC K1 (7730U / 16GB / 1TB) →
Full cluster — 3 bundles, live
02
Each manager runs its own 5-worker bundle over a dedicated LAN uplink, and all 3 managers form a full mesh with each other — every manager talks directly to every other manager, so there's no single point of failure in the control plane.
Manager (4C/4T · 32GB)
Worker (8C/16T · 16GB)
Manager mesh (full mesh)
Manager → worker uplink
3 bundles = 132 cores / 252 threads · 336 GB RAM · 18 TB SSD · 309 W total
Scaled math
03
| Component | Per bundle | × | 18-node total |
|---|---|---|---|
| Cores | 44 | 3 | 132 |
| Threads | 84 | 3 | 252 |
| RAM | 112 GB | 3 | 336 GB |
| SSD Storage | 6 TB | 3 | 18 TB |
| Power | 103 W | 3 | 309 W |
132
Cores
18TB
Storage
336GB
RAM
309W
Total Power
Notes
04
Swarm HA: 3 managers is more fault tolerance than strictly required — Swarm survives losing 1 of 3 managers already. If cost matters more than headroom, 1 manager + 15 workers gives identical compute. Keep 3 if managers are also meant to front-line mesh/DNS services per-site.