How routing to China works
What a China-bound path consists of: capacity for internal use, filing and transfer questions, and one contact on each end.
A China-bound route has three parts: capacity for your company's internal use, filing and data-transfer steps scoped with your own counsel, and one contact on each end. The path options differ in how traffic reaches mainland China and in what the provider carries.
Key facts
- Cross-border lines into mainland China generally need approval and are limited to a company's internal use under MIIT rules. ICP filing and data-transfer steps depend on your traffic and your entity; scope them with your own counsel. This is general information, not legal advice.
- A China-bound path has three parts: capacity, compliance and operations. They fail in different ways and need different owners.
- Distance sets the latency floor, hops and queuing add the rest, and the local loop is often the largest surprise.
On this page
The three parts
Every China-bound path has three parts, and they fail in different ways. Capacity is the physical and logical route: fiber, a port, a carrier backbone and the handoff at each end. You size it from the traffic profile, not from a headline bandwidth figure, because the busy-hour rate, the packet size and the latency the application can tolerate decide the design. Compliance is the set of approvals and filings around the path and the traffic: cross-border line approval, ICP filing if the site is hosted on servers or a CDN inside mainland China, and data-transfer rules that depend on what data moves and where it is stored. Operations is the human layer: one contact on each end, a named escalation path, and a change process for routes and maintenance. A well-sized circuit with no escalation path is still a business risk. The compliance primer covers the filing side.
Different teams usually own the three parts. Network engineering owns the circuit, legal and compliance own the filings, and operations owns the escalation. A quote that describes only bandwidth leaves two of the three parts to chance. Ask for the filing scope and the escalation model at the same time as the technical specification.
A China path in four segments
- Local loop
- International transit
- Border / handoff
- Domestic leg
Text description
The diagram shows four segments and no real geography. The local loop joins each site to the provider's network. The international segment crosses toward mainland China. The border handoff is where the cross-border service is defined and where approval questions attach. The domestic leg reaches the destination inside China. A route assessment names each segment and the party responsible for it; it does not publish a network map.
How a route assessment runs
-
Start with the traffic
List the sites, the applications, the busy-hour rate, and the latency and loss each application tolerates. A path sized from a bandwidth headline is a guess.
-
Map the sites and the local loops
Name each endpoint, the local loop, the handoff and the demarcation point. The local loop is often the longest lead time.
-
Compare the path options
Match each flow to transit, a private line, DIA or colocation. Write down the trade-off for each option instead of ranking products.
-
Fix the measurement plan
Agree the probe locations, the test window, the packet size and whether the result is one-way or round-trip. A single speed test is a snapshot, not a baseline.
-
Scope the filing questions with counsel
Take the public rule and your own traffic and entity to counsel. The network team does not decide the filing outcome.
Path options
Five options cover most requirements. IP transit gives full-table reachability over a BGP session and suits platforms that serve many networks. A premium China route uses engineered capacity and peering to avoid the congested default path; the quote should name the city pair and the measured figures. An IEPL or IPLC private line gives fixed capacity between two named sites and keeps traffic off the public internet. DIA or SD-WAN gives offices and branches business-grade access, with multi-path failover where the sites justify it. Colocation puts your equipment in a data center near the handoff, which can remove a long local loop. The quote should name the option, the handoff and the demarcation point. If a provider cannot say which of the five it is selling, ask again.
The options are not mutually exclusive. A platform may take transit for public traffic and a private line for database replication between two sites. An office may take DIA with an SD-WAN overlay across several countries. The right question is not which product is best but which flow needs which treatment. Write the flows down first, then map each one to an option.
What carrier-backed means
Carrier-backed means the capacity is delivered over a Chinese carrier's backbone rather than assembled from consumer internet lines. That matters for three reasons. First, the path has a defined operator and a defined handoff, so faults can be traced to a party. Second, the capacity can be ordered at a committed rate and a defined protection level instead of depending on best-effort peering. Third, the commercial and approval chain is explicit: the provider is selling capacity it can name, not capacity it hopes to route.
It does not mean the provider owns every segment. Local loops, data centers and the last mile are usually delivered with local partners, and the quote should say which parts are whose. It also does not mean the path is compliant for every use; approval and filing depend on your traffic and your entity. The MIIT entry covers the regulator, and the ICP filing and data residency entries cover the vocabulary.
Where the delay comes from
Delay has four sources. Distance sets the floor: light in fiber covers about 200,000 km per second, so a long path costs tens of milliseconds before any equipment is involved. Hops and policy add the next layer: each network boundary is a chance for a detour, a congested peer or a routing change. Queuing adds the third: a saturated port delays every packet behind it, which is why busy-hour latency differs from an idle-hour measurement. The last mile adds the fourth, and it is often the largest surprise, especially where the local loop is long or the building's wiring is old. Measure round-trip time and packet loss together; a low average with high jitter is worse for real-time traffic than a slightly higher but stable figure.
Application latency is not the same as network latency. A database write waits for a transaction round trip, a page load waits for DNS, TCP setup and several object fetches, and a video call waits for a jitter buffer. Reducing the network path helps all of them, but a badly tuned application can waste a good circuit. Measure at the application layer as well as the network layer, and fix the parts you control before buying more bandwidth.
A private line removes the filing
Myth
A private line removes the filing obligation.
Fact
A private line changes the path and the responsibility boundary. It does not change the filing obligation. Cross-border line approval, ICP filing and data-transfer rules depend on the traffic, the hosting and the entity. The public rule applies to cross-border lines and to leased international lines used for internal office purposes; how it applies to your service is a question for counsel. Read the MIIT circular (opens in a new tab) and scope it with counsel.
What to check before you buy
Ask for the city pair, the handoff type and the committed rate in writing. Ask how the provider measures latency and loss, and whether the figure is a busy-hour measurement or a best-case test. Ask what protection and diversity you get, and whether the backup shares a conduit with the primary. Ask who provides the local loop and who owns the demarcation point. Ask how route changes and maintenance are announced, and what the escalation path is at each end. Ask how cross-border approval and ICP filing are scoped, and take legal advice on your own obligations; this page is not legal advice. Finally, ask what happens at the end of the term: whether the circuit, the port and the addresses are portable, and what notice you must give.
The AS numbers guide explains how to read the network sequence in a route without over-reading it. The premium vs standard comparison puts the two transit models side by side.
Check what you learned
-
Name the three parts of a China-bound path and the owner of each.
Capacity: network engineering. Compliance: legal and compliance, with counsel. Operations: the team that owns escalation and change. A quote should cover all three.
-
The measured latency is well above the fiber floor. Where do you look next?
Check detours and peering, then queueing at the busy hour, then the local loop. Compare one-way if you can; a round-trip figure hides asymmetry.
-
A provider says the private line is compliant because it is private. What is missing?
The filing analysis. A private path changes the route, not the obligation. Ask which public rule applies and take the answer to counsel.
Answers are revealed on request. Nothing is scored, stored or sent.
FAQ
Do we need ICP filing?
Can we see a network map?
What is a point of presence?
Can you promise a specific route?
Is a private line encrypted?
Is this page legal advice?
Sources
- MIIT Circular [2017] No. 32 (via CAC): cross-border lines need approval; leased lines for internal use
- ITU-T G.114: one-way transmission time
- ITU-T Y.1541: network performance objectives for IP-based services
- ITU-T G.652: characteristics of a single-mode optical fibre and cable
- ITU-T G.654: characteristics of a cut-off shifted single-mode optical fibre and cable
Terms you'll meet
Related
- Premium vs standard transit to China
How premium China routing differs from standard international transit: what changes, what stays the same, and which questions decide the fit.
- Entering China compliance checklist
The checklist for China-bound connectivity: filing, data-transfer scoping, and the single contact and escalation path. Not legal advice.
- Autonomous system numbers, explained
What an autonomous system number identifies, how to read one on an IP result, and what it does not tell you about quality, ownership or the path.
Related products
Put this to work
Tell us the traffic, the destination and the measurement you need; we reply with the next step.