Multi-Line BGP: What It Solves, What It Does Not, and How to Verify It

# What it is: one IP, three carriers arrive via their own backbone BGP (Border Gateway Protocol) is how carriers exchange reachability. A datacenter that owns an IP block announces it via BGP to Telecom, Unicom and Mobile at once: "this prefix lives here". Each carrier's users are then guided by their own backbone to the nearest peering point — Telecom users arrive over Telecom paths, Mobile users over Mobile paths — without three physical lines or three IPs. That is the whole magic: **one IP that eliminates cross-carrier detours**. # What it solves: the old cross-carrier problem Before multi-line BGP, a single-line (Telecom-only) facility was a disaster for Unicom and Mobile visitors: inter-carrier exchange bandwidth is scarce and expensive, so cross-carrier traffic detoured and degraded badly at peak. Multi-line BGP removed that structural problem — every carrier gets a direct "own network → facility" path. For anything facing the whole consumer market (sites, APIs, games) this is **baseline, not option**: any serious node on sale today should be multi-line BGP by default. # What it does NOT solve: three separate things | Problem | Multi-line BGP fixes it? | What actually fixes it | |---|---|---| | International-segment congestion (evening-peak outbound degradation) | No | International line quality: CN2 GIA / AS9929 / CMI-friendliness | | Physical-distance latency (South China → US-West, 150 ms class) | No | Region placement (see the region hub) | | Last mile (home NAT, Wi-Fi degradation) | No | The user's own network (local three-step elimination first) | | Cross-carrier slowness inside China | ✅ Yes — its home turf | — | The most common misreading is crediting row one to BGP: "multi-line BGP optimizes international access" is marketing stitch-work. BGP solves domestic inter-carrier reach; the international segment rides whatever transit quality was purchased — a separate cost line. **A node is only as good as both layers**: domestic arrival via real BGP multi-line, international return via a quality line. Miss either and the premium bought nothing. # A short history: from dual-line dual-IP to single-IP BGP Early "dual-line" facilities ran **two physical lines and two IPs**, with "smart DNS" resolving visitors to the IP matching their carrier. The failure surface was DNS itself: cross-carrier forwarding by local resolvers, patchy EDNS Client Subnet support, inconsistent cache expiry — a steady share of users landed on the wrong line, and troubleshooting was a nightmare ("the site opens for me but not for my father" is usually this). BGP multi-line took the decision away from DNS and gave it back to routing: one IP, each carrier's backbone delivers its own users to the nearest entry. No resolution ambiguity, no stale cache. This history is why "smart-DNS dual-line" remains shorthand for low-end, and real BGP multi-line deserves its premium. # Verify a node is genuinely multi-line BGP ```bash # 1) run one MTR from each carrier side (Telecom/Unicom/Mobile exits; a phone hotspot = Mobile) mtr -rwbzc 50 # 2) read each path: each carrier should arrive via its own backbone, not detour through a mid-path exchange # 3) check the IP's BGP origins (bgp.he.net online, or RADB from the shell) whois -h whois.radb.net -- '-i origin' | grep # 4) reverse check: traceroute from the server to each carrier's test point; the return path should be equally direct ``` Fake multi-line betrays itself in the paths: if all three carriers' MTRs converge on the same inter-carrier exchange mid-path, the "multi-line" claim is copywriting. The full method (time slots, sample sizes, archiving) lives in the diagnostics hub. # Two layers, never stitch them Repeat the distinction that matters most: **BGP multi-line is an access-layer property** — how the three carriers get in the door. **Line quality (CN2 GIA / AS9929 / CMI) is a transport-layer property** — how the international segment travels once inside. Sales copy fuses them ("CN2 GIA multi-line BGP!") into something that sounds doubly premium but is two independent dimensions. Read a config sheet with two separate questions: "is domestic arrival real BGP multi-line?" and "what transit quality does the international segment ride?" Only when both answers are explicit does a premium become spendable. # Who should pay the premium - **Consumer-facing services (e-commerce, content, games):** yes — this is floor configuration. - **Enterprise intranet / private-line scenarios:** not necessarily — a single-carrier user base is cheaper on single-line plus same-carrier peer. - **Outbound (international) businesses:** BGP multi-line is only the passing grade; put the budget into international line quality (CN2 GIA / CMI-friendly nodes) and region placement. The full decision tree lives in the routing hub. Run the four verification actions against your candidates, then match node × line in the live catalogue. One sentence to keep: **BGP multi-line owns the door, line quality owns the sea, region placement owns the distance — three layers, none substitutable.**