01Three pain signals: coast labels hide routing
Stable CPU charts hide RTT, RAM, and disk debt that only shows up under API load.
- RTT blind spots: Ping games skip us-east-1, us-west-2, and your production SaaS edge.
- Sixteen gigabyte wall: Two OPENCLAW_HOME trees plus Simulator look like node instability.
- Port and SSD debt: 18789 collisions and dual-gateway logs fill one terabyte before CPU turns red.
02US East vs US West: node use-case decision matrix
From one production egress, same DNS, same VPN SKU: map workload to region—not to marketing labels like «East Coast» or «West Coast».
| Workload | US East (Virginia) | US West (LA / San Jose) | Quick call |
|---|---|---|---|
| AWS us-east-1 API | Data plane often optimal | Transcontinental higher | Data in East → US East |
| us-west-2 / GitHub CI | Usable, rarely top | Open-source path often shorter | West CI → US West |
| OpenClaw always-on gateway | East-coast clients | APAC night batch | Pick from median lights |
| APAC daytime VNC | APAC node or parallel slot for UI | See APAC matrix articles | |
03RTT thresholds: AWS us-east-1 and us-west-2
Per region run ten HTTPS probes from the same exit. Log median and p95 in the release change log. Jitter above fifteen milliseconds for five minutes: throttle push intervals.
| Probe target | US East node | US West node | Traffic light |
|---|---|---|---|
| us-east-1 regional endpoint | Median often under 50 ms | Often above 80 ms | Median above 70 ms → yellow |
| us-west-2 regional endpoint | Transcontinental | Median often under 45 ms | Median above 90 ms → red |
| REST / WebSocket SaaS | Judge p95, not single spikes; three failed runs → switch region | Security review if MITM suspected | |
04OpenClaw multi-instance: OPENCLAW_HOME and port 18789
- Primary: baseline 18789, dedicated
OPENCLAW_HOME, dedicated LaunchAgent. - Replica: port offset +20 / +40; four-column registry (project · port · HOME · agent).
- Verify:
lsof -i :18789, health probe, SSH tunnel on the same port.
Do not share HOME trees between US East and US West—sync fights cost more than rent savings. After each offset run openclaw doctor; use launchctl print to map PIDs before buying a second parallel slot.
05Mac mini M4: sixteen gigabyte multi-session vs twenty-four gigabytes
| Scenario | 16 GB | 24 GB | Call |
|---|---|---|---|
| Two tabs + Simulator + one gateway | Baseline OK | Headroom | Single region OK |
| +1 always-on OpenClaw session | About +2 GB RAM | +1 session possible | Throttle or queue |
| Two gateways twenty-four-seven | Swap risk | Testable | Second parallel slot |
06Parallel slots and 1TB/2TB expansion cost thresholds
| Check | 1 TB | 2 TB | Rent term hint |
|---|---|---|---|
| Free space | Two-week buffer above fifteen percent | Dual HOME logs under fifteen percent free | Daily: dual-region test |
| Artifacts | Single OpenClaw line | Dual HOME + model cache | Weekly: two-to-four week sprint |
| Parallel slot | US East API + US West build → two neokvm instances | Monthly: always-on gateway | |
Confirm live rows on pricing. Checkout: purchase, US West hub: US West nodes.
07TestFlight / App Store Connect: upload selection tips
- Large IPAs: median often favors US West—see TestFlight FAQ.
- API metadata in us-east-1: with a green light use US East; split Archive vs Transporter across parallel slots.
- FAQ: Is US East always slower for upload? No—only your ten medians count.
08Install path: install.sh and Node 24+ quick reference
- One-liner:
curl -fsSL …/install.sh | bash— needs Node 24+ and curl-capable egress. - Check:
node -v,openclaw doctor; on failure inspect DNS and Homebrew mirrors. - onboard: never share
~/.openclawbetween primary and replica instances.
Ops depth: gateway troubleshoot guide.
09Five ops steps and citable facts
- 1. Dual-region × ten runs toward AWS and target API—log green, yellow, red.
- 2. Apply decision matrix; write region into change log.
- 3. Draw OPENCLAW_HOME and port registry.
- 4. Walk 1TB/2TB checklist and parallel slot with finance; checkout by region.
- 5.
openclaw doctorOK—console; questions: help.
US East or US West for OpenClaw? Let ten medians to AWS and your SaaS API decide—not coast marketing. Two gateways on sixteen gigabytes? Budget a parallel slot or twenty-four gigabytes. Split regions? US East for us-east-1 data plane, US West for GitHub CI and large IPA uploads when your lights differ.
| Stability signal | Green | Escalation |
|---|---|---|
| API success rate | ≥ 98% over ten probes | Two red runs → switch region |
| Gateway health | 18789 probe returns 200 OK | Port collision or wrong SSH tunnel |
| Disk headroom | Above fifteen percent free before release week | Below ten percent during chunked upload |
Lock your neokvm node with the decision matrix
City FAQ: Los Angeles vs San Jose. Home, purchase, US West nodes, pricing.