When clients ask us "what latency can you do on Shanghai-Tokyo," the answer is 26ms. That number isn't arbitrary — it's constrained by physics. Understanding the physical limits of latency helps you evaluate whether a provider's claims are credible, and helps you plan your business architecture. This article starts from basic physics to explain why latency is what it is, and why IPLC leased lines are far faster than the public internet.
Speed of Light in Fiber
All network data ultimately travels as light signals through optical fiber. The speed of light in vacuum is c = 299,792 km/s, but fiber isn't vacuum — it's made from high-purity silica glass with a refractive index of approximately 1.468. The effective speed of light in fiber is about 68% of vacuum light speed:
v = c × 0.68 ≈ 204,000 km/s
To put it another way: each kilometer of fiber adds about 4.9 microseconds (μs) of one-way propagation delay, or roughly 4.9ms per 1,000 km. This is an unbreakable physical floor — no matter how expensive your equipment or how premium your fiber, light simply travels at this speed.
From Distance to Latency: Theoretical Calculations
Knowing the speed of light in fiber, we can calculate the theoretical minimum latency between any two points. Note that we're discussing one-way delay here; actual ping values (RTT, round-trip time) are double this number. Also, fiber routes are never perfectly straight — submarine cables route around ocean trenches and continental shelves, while land cables follow railways and highways. Actual fiber distance is typically 1.2-1.5x the straight-line distance.
Here are theoretical delays and Areapac measured values for several typical routes:
Guangdong — Hong Kong
Straight-line distance: ~30km, Fiber distance: ~50km
Theoretical one-way delay: 0.25ms, Theoretical RTT: 0.5ms
Areapac measured RTT: 2.4ms
Why the gap: Despite the extremely short physical distance, signals pass through multiple network devices (routers, switches, optical amplifiers), each adding a few hundred microseconds of processing delay. The 2.4ms includes processing time from approximately 4-6 devices.
Shanghai — Tokyo
Straight-line distance: ~1,750km, Fiber distance: ~1,800km (via submarine cable)
Theoretical one-way delay: 8.8ms, Theoretical RTT: 17.6ms
Areapac measured RTT: 26ms
Why the gap: The submarine cable departs from the Chongming landing station in Shanghai, crosses the East China Sea to the Chikura or Maruyama landing station in Japan, then connects to a Tokyo data center. The path is about 3-5% longer than a straight line, and landing station equipment plus routing equipment at both data centers add approximately 8ms. At 26ms, this is very close to the physical limit.
Shanghai — Seoul
Straight-line distance: ~870km, Fiber distance: ~1,050km (via Yellow Sea submarine cable)
Theoretical one-way delay: 5.1ms, Theoretical RTT: 10.3ms
Areapac measured RTT: 21ms
Why the gap: The submarine cable route detours near Qingdao, making the actual fiber path about 20% longer than the straight line. Including equipment delay at both ends, 21ms is an excellent result from an engineering perspective.
Beijing — Frankfurt
Straight-line distance: ~7,400km, Fiber distance: ~7,500km (via land cable through Russia/Kazakhstan)
Theoretical one-way delay: 36.7ms, Theoretical RTT: 73.5ms
Areapac measured RTT: 113ms
Why the gap: Intercontinental cable routes are winding, with actual fiber distance potentially reaching 9,000-10,000km. The path passes through multiple repeater stations and OADM (Optical Add-Drop Multiplexer) nodes, each introducing additional delay. 113ms for a Eurasian land cable route is industry-leading.
Why Public Internet Latency is 2-5x Higher
If you ping a Tokyo server from Shanghai on regular broadband, latency is typically 60-150ms or even higher. Same physical distance — so why is public internet latency far higher than IPLC's 26ms? Three reasons:
1. Too Many Routing Hops
A public internet packet leaving your computer traverses your local switch, metro core router, provincial backbone router, international gateway router, then the destination carrier's gateway, backbone, metro, and access networks... potentially 15-25 routing devices in total. Each router adds 0.2-2ms of lookup and forwarding delay, meaning device processing alone can add 10-30ms.
2. Congestion Queuing
Public internet links are shared. During peak hours (evenings, holidays), international gateway bandwidth may be saturated by video streaming traffic. When a router's input rate exceeds its output rate, packets queue in buffers waiting their turn. This queuing delay is unpredictable, fluctuating from 0ms to several hundred ms — which is why public internet ping values often "spike."
3. Routing Inefficiency
Public internet BGP routing doesn't optimize for lowest latency — it's based on commercial relationships between carriers (peering/transit agreements). Your packet might route from Shanghai to Guangzhou before heading to Tokyo, or from Beijing via the US West Coast back to Tokyo. This kind of "detour" is extremely common on the public internet.
Why IPLC Gets Close to Physical Limits
IPLC (International Private Leased Circuit) achieves near-theoretical latency because of three design principles:
Dedicated fiber channels: IPLC allocates a dedicated wavelength or timeslot for you on the carrier's optical transport network. Your data doesn't compete with other users' traffic for bandwidth, completely eliminating congestion queuing delay.
Minimum hops: IPLC is a point-to-point connection — from landing station A directly to landing station B, passing through only necessary optical amplifiers and OADM nodes, with no IP-layer routing forwarding. Equipment delay is minimized.
Optimized routing paths: IPLC uses carefully selected shortest-path cable routes. Shanghai to Tokyo goes via a direct East China Sea submarine cable, rather than detouring through other cities.
Areapac's Latency Performance
All Areapac leased line products run on Tier-1 carrier optical transport networks using dedicated channels (dedicated wavelength/timeslot). This means our latency numbers approach the physical limit for each route:
- Guangdong — Hong Kong: 2.4ms RTT
- Shanghai — Tokyo: 26ms RTT
- Shanghai — Seoul: 21ms RTT
- Beijing — Frankfurt: 113ms RTT
These aren't "best case possible" marketing numbers — they're the stable latency values our customers actually measure every day in production environments. The key to low latency isn't how advanced the equipment is, but how direct the route is, how few hops there are, and whether bandwidth is dedicated. These are the core values of a leased line.
One-Line Summary
Latency = distance ÷ speed of light + equipment delay. IPLC leased lines compress total latency to near the physical limit by eliminating congestion, reducing hops, and optimizing routes. The public internet can't do this because it's inherently shared and multi-hop.