Network Latency Calculator: Fiber Delay & File Transfer Time
Network Latency & Data Transfer Time Calculator
Propagation delay over fiber, copper or radio from a distance or two cities, serialization and per-hop delay, then how long a file transfer really takes with protocol overhead and a TCP window limit. Everything recalculates as you type; the URL keeps your inputs so you can share them.
1. Distance & propagation delay
1.0 = cable follows the straight great-circle line. Real fiber follows roads, rail and sea-cable landings, so it is longer: enter e.g. 1.3 for a 30% longer route.
Latitude, longitude in decimal degrees (north and east positive). Pick a city or type your own coordinates.
Same path in other media
| Medium | Delay per km | One-way | RTT |
|---|
2. Serialization, hops & total latency
Header 14 + FCS 4 + preamble/SFD 8 + inter-frame gap 12 bytes. Untick if the size already is the full frame on the wire.
Your estimate. It is near zero on an idle network and grows with congestion and buffer size.
3. Data transfer time
Bits per second, decimal (1 Gbps = 10^9 bit/s).
What this network latency calculator does
Latency between two sites has a hard floor set by physics: the signal has to travel the distance at the speed of light in the medium. This calculator works out that floor for fiber, hollow-core fiber, copper, coax and radio, from a distance you enter or from two cities or coordinates (great-circle distance). It then adds what the network itself contributes, the serialization delay of a packet at each hop and your estimate of queuing and processing, and finally turns latency and bandwidth into a realistic data transfer time. The transfer part counts Ethernet, IP and TCP header overhead and the TCP window limit, which is why a 1 Gbps link across the Atlantic often does not move a file at 1 Gbps.
How it works
Propagation delay
one-way delay = path length x n / c, with c = 299,792.458 km/s (exact by the SI definition of the metre) and n the group index of the medium. Copper and coax data sheets give a velocity factor (VF or NVP) instead, so there n = 1 / VF. RTT is twice the one-way delay.
| Medium | Value used | Delay per km | Source |
|---|---|---|---|
| Single-mode fiber, 1550 nm | n = 1.4682 | 4.897 µs | Corning SMF-28 Ultra product information sheet PI-1424 (effective group index 1.4682 at 1550 nm) |
| Single-mode fiber, 1310 nm | n = 1.4676 | 4.895 µs | Corning PI-1424 (1.4676 at 1310 nm) |
| Hollow-core fiber | n ≈ 1.001 | 3.339 µs | Photonica, “Hollow-core fiber explained” (group index about 1.001, 3.34 µs/km); arXiv 2607.06877 (about 3.3 µs/km) |
| Cat6 twisted pair | NVP 67% | 4.979 µs | Siemon Solution 6 UTP cable spec sheet (NVP 67%); other Cat6 cables differ by a few percent |
| Coax RG-6 (foam) | VF 85% | 3.924 µs | Priority Wire RG-6/U data sheet (velocity of propagation 85%) |
| Microwave / radio in air | n ≈ 1.000315 | 3.337 µs | ITU-R P.453 (average sea-level refractivity N0 = 315) |
| Vacuum / free space | n = 1 | 3.336 µs | SI definition of c |
A handy rule of thumb follows from the fiber row: about 4.9 µs per km, so roughly 1 ms of RTT for every 100 km of fiber path.
Great-circle distance and route factor
For two locations the tool uses the haversine formula on a sphere with the IUGG mean Earth radius R1 = 6371.0088 km: a = sin²(Δφ/2) + cos φ1 cos φ2 sin²(Δλ/2), d = 2R asin(√a). A sphere differs from the WGS 84 ellipsoid by up to about 0.5%, which is far smaller than the route error. City presets are city-centre coordinates from OpenStreetMap (Nominatim), rounded to 4 decimals.
The great-circle line is the shortest possible path, and real cables rarely follow it. Terrestrial fiber runs along roads, railways and pipelines; submarine systems go via landing stations and around coastlines. The route factor multiplies the distance to account for that. Leave it at 1.0 when you enter a measured fiber route length, or when you want the theoretical minimum. To learn your real factor, divide a measured RTT (minus equipment delay) by the great-circle RTT this tool shows.
Serialization, queuing and total latency
serialization = bits on the wire / link rate. A 1500-byte IP packet on Ethernet occupies 1538 bytes of wire time: 14-byte header, 4-byte FCS, 7-byte preamble, 1-byte SFD and 12-byte inter-frame gap, the same accounting RFC 6349 (section 4.1.1) uses. A store-and-forward device has to receive the whole frame before it sends it on, so serialization is paid once per hop (cut-through switches avoid most of it). The estimate is one-way = propagation + hops x (serialization + queuing/processing) and RTT = 2 x one-way. Queuing depends on load and buffer depth and cannot be calculated from the topology, so it is your input.
Data transfer time
time = bytes x 8 / throughput, and throughput = min(link rate x efficiency, TCP window / RTT). Decimal units (KB, MB, GB, TB = powers of 1000) and binary units (KiB, MiB, GiB, TiB = powers of 1024, IEC 80000-13) are kept separate because a “1 TB” disk and a 1 TiB file differ by almost 10%. Link rates are always decimal: 1 Gbps = 109 bit/s.
Protocol efficiency follows RFC 6349 section 4.1.1: TCP payload per frame divided by bytes on the wire. For a 1500-byte MTU, IPv4 and TCP without options: (1500 - 20 - 20) / (1500 + 38) = 1460 / 1538 = 94.93%, so 1 Gbps carries at most about 949 Mbit/s of TCP payload; RFC 6349 gets the same 94.9 Mbps for Fast Ethernet. With the TCP timestamps option (12 bytes, on by default in Linux) it is 1448 / 1538 = 94.15%, and with IPv6 1440 / 1538 = 93.63%. A 9000-byte jumbo MTU raises it to 8960 / 9038 = 99.14%.
The window limit comes from the fact that a TCP sender can only have one window of unacknowledged data in flight per round trip, so a single flow cannot go faster than window / RTT. The window needed to fill a link is the bandwidth-delay product, BDP = link rate x RTT (RFC 6349 sections 3.3.1 and 4.1.2). Windows above 64 KiB need the window scale option from RFC 7323. The model ignores slow start, packet loss and the connection handshake, so treat the result as a best case for one TCP flow; the actual window depends on the operating system's buffer auto-tuning, not on the router vendor.
Worked examples
- 1000 km of single-mode fiber. 1000 x 1.4682 / 299,792.458 = 4.897 ms one-way, 9.795 ms RTT. Over the same 1000 km in vacuum or by microwave it would be 3.336 ms one-way, about 32% less.
- London to New York. The great-circle distance is 5,570 km. In fiber that is 27.28 ms one-way and 54.56 ms RTT at best. If the cable route is 30% longer (route factor 1.3), the RTT is 70.93 ms. Frankfurt to Warsaw (890 km) gives a minimum of 8.72 ms RTT, and London to Singapore (10,847 km) gives 106.25 ms.
- Serialization at different speeds. A 1500-byte packet is 1538 bytes on Ethernet: 12.304 µs at 1 Gbps, 1.230 µs at 10 Gbps, but 1.230 ms at 10 Mbps. Over 10 hops at 1 Gbps with 0.05 ms of queuing and processing per hop, London to New York adds 0.623 ms one-way, so the total RTT estimate becomes 55.81 ms.
- 10 GB over 1 Gbps, no window limit. 10 x 109 x 8 = 8 x 1010 bits. With 94.93% efficiency the goodput is 949.3 Mbit/s, so the copy takes 84.27 s (1 min 24.3 s), not the 80 s you get from the raw link rate.
- The same 10 GB from London to New York with a 4 MiB window. 4 MiB = 4,194,304 bytes; at 55.81 ms RTT that allows 4,194,304 x 8 / 0.05581 = 601.3 Mbit/s. The window, not the link, is the bottleneck, and the transfer takes about 2 min 13 s. Filling 1 Gbps would need a window of at least the BDP, about 7 MB.
- Old 64 KiB window. Without window scaling (65,535-byte window) and a 70 ms RTT, one TCP flow is capped at 65,535 x 8 / 0.07 = 7.49 Mbit/s, no matter how fast the link is.
FAQ
How much latency does fiber add per kilometer?
About 4.9 microseconds per km one way. Light in standard single-mode fiber travels at c divided by the group index (1.4682 at 1550 nm for Corning SMF-28 Ultra), which is roughly 204,000 km/s. That makes about 1 ms of round-trip time for every 100 km of fiber route.
Why is my measured ping higher than the calculated latency?
The calculator gives the physical minimum for the distance you enter. Real cables are longer than the great-circle line, and every router, switch, firewall and server adds serialization, queuing and processing delay. Use the route factor and the per-hop inputs to model those, or divide your measured RTT by the great-circle RTT to see the effective route factor.
Is microwave or hollow-core fiber really faster than normal fiber?
Yes. Radio through air and light in a hollow air core both travel at almost the vacuum speed of light, about 3.34 microseconds per km, against about 4.9 microseconds per km in solid glass fiber. Over the same path length that is roughly 32% less delay, which is why trading firms use microwave links and hollow-core fiber.
How do I calculate data transfer time?
Divide the data size in bits by the effective throughput. Effective throughput is the link rate times the protocol efficiency, but never more than the TCP window divided by the RTT. Example: 10 GB is 80,000,000,000 bits; at 1 Gbps with 94.93% efficiency that is 949.3 Mbit/s, so the transfer takes 84.27 seconds.
What is the maximum TCP throughput on a 1 Gbps Ethernet link?
About 949 Mbit/s with a 1500-byte MTU, IPv4 and TCP without options. Each 1460-byte payload needs 1538 bytes on the wire (IP and TCP headers, Ethernet header and FCS, preamble, SFD and inter-frame gap), and 1460 / 1538 = 94.93%. RFC 6349 uses the same calculation.
Why does latency limit the speed of a file transfer?
A single TCP connection can only send one window of data before it has to wait for acknowledgements, which take one round trip. Throughput is therefore at most window size divided by RTT. A 4 MiB window over a 55.8 ms RTT allows about 601 Mbit/s, even on a 1 Gbps link. Larger windows (window scaling), lower RTT or parallel streams fix this.
Is 1 GB 1000 MB or 1024 MB?
In SI units 1 GB is 1,000,000,000 bytes and 1 GiB is 1,073,741,824 bytes (1024 x 1024 x 1024). Network speeds always use decimal prefixes, so 1 Gbps is 1,000,000,000 bits per second. The calculator lets you pick decimal or binary size units explicitly.
Comments
Post a Comment