Latency is the delay between sending data and its arrival. It is measured one way, from sender to receiver, or as round-trip time, which is what a ping shows. In a chat app it is how long a message takes to appear on the other device; in a call it is the mouth-to-ear delay for audio or the glass-to-glass delay for video.

What is a good latency for a video call?

For voice, ITU-T G.114 recommends keeping one-way mouth-to-ear delay under 150 milliseconds; conversation gets awkward above 250 to 300 milliseconds and breaks down beyond 400. For video, a glass-to-glass delay of 200 to 400 milliseconds is typical of good conferencing systems. Network round-trip time is only part of that figure; encoding, jitter buffering and decoding add the rest.

What is the difference between latency and bandwidth?

Bandwidth is how much data a link can carry per second; latency is how long each piece of data takes to get there. A satellite link can have high bandwidth and very high latency; a nearby server on a slow link has low latency and low bandwidth. Real-time chat and calls need low latency far more than high bandwidth, since a voice stream uses only tens of kilobits per second.

What is the difference between latency and jitter?

Latency is the average delay; jitter is how much that delay varies from packet to packet. A call with a steady 100 millisecond latency sounds fine. A call whose packets arrive anywhere between 60 and 180 milliseconds sounds choppy unless a jitter buffer holds packets back to smooth them out, which adds latency. Stable latency is worth more than low but erratic latency.

How do you keep latency low in a chat app?

Push messages over a persistent connection such as WebSocket or XMPP instead of polling, run servers in the same region as the users, keep the number of hops between services small and send media over UDP so lost packets do not stall the rest. Ethora follows this model: XMPP for chat, WebRTC with an SFU for calls, and dedicated or self-hosted deployment in the region you choose.

Glossary

Latency

The wait between send and arrive. Latency decides whether a chat feels instant and whether a call feels like a conversation, and only part of it is the network’s fault.

General definition

Latency is the delay between something being sent and it arriving. Network engineers measure it one way (sender to receiver) or as round-trip time (RTT), the time for a packet to go out and an acknowledgement to come back, which is what ping reports. In a chat product it is the gap between one person tapping send and the other person’s screen updating. In a call it is mouth-to-ear delay for audio and glass-to-glass delay for video: from light hitting one camera to pixels appearing on the other screen.

Latency has several sources, and only some of them are the network’s fault. Propagation delay is physics: light in fibre covers roughly 200 kilometres per millisecond, so a transatlantic round trip cannot go much below 60 to 70 milliseconds. Queueing delay comes from congested routers and buffers. Processing delay comes from every server hop, TLS handshake and database write on the path. For media, encoding, the jitter buffer that smooths out jitter, decoding and rendering each add tens of milliseconds. Bandwidth is a different quantity: a link can have high throughput and still be slow to respond.

Voice: ITU-T G.114 recommends one-way mouth-to-ear delay under 150 milliseconds; conversation degrades noticeably above 250 to 300 and turn-taking breaks down beyond 400
Video calls: glass-to-glass delay of 200 to 400 milliseconds is typical of good conferencing; going lower needs specialised tuning
Chat: a message delivered in under 200 milliseconds feels instant; typing indicators and presence feel wrong if they lag by more than a second
Streaming: broadcast HLS runs seconds behind live, which is fine for a webinar and useless for a conversation

Reducing latency is mostly about removing hops and waiting. Real-time transports such as WebSocket keep a connection open so a message is pushed the moment it exists, instead of waiting for the next poll. Media runs over UDP so a lost packet does not stall the ones behind it. Servers placed in the same region as the users, an SFU instead of a mixing server, and adaptive bitrate that avoids congestion all shorten the path. What cannot be engineered away is distance, which is why region choice matters.

Prefer to watch? Latency explained in about two minutes.

In the Ethora ecosystem

Ethora runs chat over persistent connections to an XMPP server, so messages, typing indicators, presence and read receipts are pushed as they happen rather than polled, and calls use WebRTC with an SFU that forwards media without the decoding and mixing delay of an MCU. Those are the two architectural choices that keep application-level latency low; the rest is geography.

Geography is the part you control. A dedicated Ethora server runs in the cloud region you choose, inside your own AWS, DigitalOcean or other account, and a self-hosted deployment goes wherever you put it, so an EU customer base is served from the EU and a North American one from North America. Keeping chat, media and AI services in the same region also keeps the hops between them short, which matters when an AI agent replies inside a room or a call is started from a chat. Load-testing tools ship with the self-hosted stack so you can measure delivery times under your own expected concurrency before launch.

Latency

General definition

In the Ethora ecosystem

Ship chat features, not chat infrastructure

Talk to our team

Pick a time that works for you