Skip to content

What is latency?

Latency is the time between sending a request and receiving its answer, measured in milliseconds (ms). Uptime tells you whether a system answered; latency tells you how long the answer took — and for the person waiting at the screen, the second number is usually the one that shapes the experience.

The millisecond scale

A few reference points calibrate the units:

  • 1–10 ms — the same building or the same city. A ping between two servers in one data center lands here.
  • 20–80 ms — within the same country or region, or to a nearby cloud region.
  • 100–300 ms — crossing continents. Physics sets the floor: no upgrade brings a server in Europe closer than the speed of light allows.
  • Server processing time stacks on top of all of this — a query-heavy page can add hundreds of milliseconds over a perfectly fast network.

What users perceive is well established: below roughly 100 ms an interaction feels instant; around one second there's a noticeable pause; past a few seconds users assume the system is broken and start reloading or leaving. UpAll's checks draw the final line at a fixed 10 seconds — beyond that, a check doesn't count as slow, it counts as failed.

Latency is often confused with bandwidth, which is a different dimension entirely — see latency vs bandwidth.

Network latency vs server latency

The number an outside check records is the sum of two very different things:

  • Network latency — the round trip across the wire: distance, routing hops, congestion. Determined mostly by geography and the quality of the path.
  • Server latency — what happens after the request arrives: queueing, application code, database queries.

A practical way to separate them is to run two monitors against the same host. A Ping monitor measures the near-pure network round trip (how Ping checks work), while an HTTP monitor measures the network plus the server's work. If Ping stays flat while HTTP climbs, the problem lives in the application, not the line.

Why "up but slow" matters

A server answering in four seconds passes every availability test and fails every user. Slowness costs the same things downtime costs — abandoned carts, frustrated staff, support calls — while looking perfectly green on a naive dashboard.

That's why UpAll treats latency as a health signal, not just a chart. Each monitor can carry a degraded threshold — UpAll recommends a default per check type (2,000 ms for a website, 500 ms for a TCP service, 100 ms for a ping check), and a Service class picker lets you pick a different one or set a custom number: responses slower than it flip the monitor to an amber Degraded state — visibly unhealthy without opening an incident (degraded latency threshold). Every check's response time is recorded and drawn on the monitor's latency chart (reading monitor charts), and summarized as p95 on status pages and in monthly reports — see percentiles and p95 for why that summary beats an average.