> ## Documentation Index
> Fetch the complete documentation index at: https://docs.daily.co/llms.txt
> Use this file to discover all available pages before exploring further.
# Global Mesh Network (Mesh SFU)
> Daily's global infrastructure for delivering real-time video and audio to anyone, anywhere in the world.
## Overview
Daily’s **Mesh SFU** (Selective Forwarding Unit) is a network architecture in which our globally distributed media servers are used to reduce the impact of the public internet and isolate it to the end user’s local ISP. In many cases, this reduces the latency, packet loss, and jitter — improving the overall call experience for end users around the globe. Some of our competitors call this architecture Cascaded Media Servers.
## How does Daily's Mesh SFU work?
It’s best to start with an example — if a user in California has a video call with a user from New York, the traditional approach would be for both of them to connect to a centralized media server (an SFU). This central media server can be equidistant to both users, or close to one user and far from the other. In either case, the connection between the end user and the SFU is on the public internet and, therefore, subject to disruption and unpredictability of the public internet.
This is remedied by the Mesh SFU topology, in which the end user is directly communicating with their nearest SFU, **and those SFUs communicate with each other over our provider’s internal backhaul network,** avoiding as much exposure to the public internet as possible. Our data shows that the backhaul connection within a cloud infrastructure is far more performant compared to the connections between the ISPs.
## AWS Infrastructure
Our Mesh SFU topology is built on [AWS infrastructure](https://aws.amazon.com/about-aws/global-infrastructure/), which has 400+ POPs (points of presence) between 26 regions around the world. As AWS continues to expand its availability zones and POPs across the globe, Daily’s own capacity and performance also scale alongside it.
The AWS 50th-percentile ping benchmarks below demonstrate very low latencies between AWS regions within both the EU and US. End user latencies would be even lower than these averages, since `user <-> SFU` distance is generally smaller than `SFU <-> SFU` distance.
| | EU Frankfurt | EU Stockholm | EU Milan | EU Ireland | EU London | EU Paris | |
| ------------ | ------------ | ------------ | -------- | ---------- | --------- | -------- | ---- |
| EU Frankfurt | 3 | 23 | 12 | 26 | 17 | 12 | |
| EU Stockholm | 22 | 3 | 32 | 42 | 33 | 31 | |
| EU Milan | 12 | 32 | 2 | 35 | 26 | 20 | |
| EU Ireland | 27 | 43 | 35 | 2 | 13 | 19 | |
| EU London | 18 | 34 | 27 | 13 | 4 | 11 | |
| EU Paris | 13 | 32 | 21 | 19 | 11 | 4 | |
| Average | | | | | | | 20ms |
| | US Virginia | US Ohio | US California | US Oregon | |
| ------------- | ----------- | ------- | ------------- | --------- | ---- |
| US Virginia | 5 | 14 | 63 | 68 | |
| US Ohio | 14 | 4 | 53 | 55 | |
| US California | 62 | 52 | 4 | 22 | |
| US Oregon | 68 | 54 | 22 | 2 | |
| Average | | | | | 35ms |
## Further reading
A detailed look at the architecture behind Daily's mesh SFU infrastructure — how media routing, region selection, and backbone networking work under the hood.