Regions & Datacenters: Choosing Between HK, JP, SG and US-West
Latency physics, per-region line availability, compliance differences, soft metrics (SLA and remote hands), citation rules for measured data, and a workload-b
# Translate "where are my users" into "where is my box"
Region choice mistakes cannot be rescued by any line budget. This hub maps latency characteristics, line availability, compliance differences and typical uses per region, with citation rules for measured data.
## Quick decision block
- South-China users, lowest latency → **HK**
- East/North-China users, gaming and trade → **JP**
- Southeast-Asia market, APAC balance → **SG**
- North-America market, budget-sensitive, big bandwidth → **US-West**
## Four regions in one table
| Region | Experience latency from mainland China | Line resources | Filing | Typical use |
|---|---|---|---|---|
| HK | South 10–40 ms · national 30–60 ms | CN2 GIA, CMI, fullest local lines | none | trade sites, low-latency CN-facing business |
| JP | East 30–60 ms · North 50–80 ms | SoftBank, NTT, KDDI, CN2 | none | game servers, East-CN audiences, APAC relay |
| SG | South 60–90 ms · SEA local 5–20 ms | CN2, CMI, international mix | none | SEA markets, cross-border APIs, distribution |
| US-West | 130–180 ms (physics floor) | CN2 GIA, CMIN2, standard mix | none | NA users, big storage/bandwidth, cost-sensitive |
## The physics floor of latency
Light sets the floor: ~200,000 km/s in fibre, plus hop and device processing. Shanghai → Los Angeles has a one-way physical limit near 60–70 ms; a 130 ms round trip is already excellent. Any "US-West ping 80 ms" claim is false on its face. Accept physics first, then optimize: what line quality changes is peak degradation, never distance.
## Line availability per region
- **HK:** CN2 GIA, CMI, PCCW/HGC local lines — the fullest tri-carrier optimization, and the highest prices across the board.
- **JP:** SoftBank, NTT, KDDI, CN2 GIA; Unicom users' experience generally beats Telecom's standard lines.
- **SG:** CN2, CMI, NTT/PCCW international mix; for SEA-local access, weigh local peering first.
- **US-West:** CN2 GIA, CMIN2 (Mobile premium), standard 163/Unicom/NTT mix; evening-peak spread is huge — always read the return path.
Line background and verification in the routing hub; real numbers for your users' direction via the speed test hub per region.
## Compliance and operating differences
Mainland China: ICP filing plus real-name, assisted by the access provider. HK/JP/SG/US: no filing, live on provision — but content-compliance liability stays with the operator (infringement, fraud and spam get actioned everywhere). Data-sovereignty-sensitive workloads (finance/medical/government) must confirm data-residency requirements before choosing. Also check provider AUPs: PBNs and mail marketing face very different tolerances across regions and vendors.
## Soft metrics: SLA and remote hands
When same-spec quotes differ, the gap is usually soft metrics: power/network availability commitments (parallel maintenance only credible at Tier III+), remote-hands response speed (reboots, reseats), ticket quality. Dedicated and PBN businesses live on remote hands — a dead box nobody presses restart for makes every line choice moot. Checklist and reading method in Academy lesson 4.
## Citation rules for measured data
Latencies on this page are **experience ranges, not promises**: real numbers move with time slot, source carrier and routing. Every measured figure cited on this site carries test time, source region+carrier, target datacenter, tool and sample size — record the same five when you self-test, or your numbers are not comparable (method in the diagnostics hub).
## Workload × region decision table
| Workload | Main users | First choice | Second | Key consideration |
|---|---|---|---|---|
| Trade site | CN + overseas mix | HK | SG | CN reachability vs overseas balance |
| Pure overseas e-commerce | EU/US | US-West | EU | bandwidth cost + local payment callbacks |
| Game server (Chinese-speaking) | East/South CN | JP | HK | jitter first, dedicated CPU |
| SEA local business | SEA | SG | MY | local peering + data residency |
| Backup/DR | — | different region from prod | US-West | fault-domain isolation + capacity cost |
## Dual deployment logic
"Business in CN, servers overseas" is the commonest combo: no filing, fast provision, content autonomy — priced in return-path line quality, which is exactly why HK/JP + CN2 GIA/CMI exists. The reverse ("users overseas, box in CN") almost never works: filing constraints plus egress cost make it lose twice. One-line criterion: **servers follow whoever calls the service logic; content follows users (CDN solves it).** Database and API in HK, static worldwide via CDN — the standard architecture for small-and-medium outbound business.
## Three procurement practicalities
- **Currency:** overseas facilities price in USD; budget for FX float in your bookkeeping currency.
- **Billing cycle:** monthly for exploration, yearly only after three months of stable load (yearly discounts trade flexibility for price).
- **Growth path:** start one tier below your forecast; vertical resize is minutes, region migration is a project.