Agent latency as a server capacity signal

Most VICIdial admins treat Latency complaints as a network problem and hand them off to the agent's IT department. That is the right first step, but it misses a pattern that appears repeatedly on busy systems: Agent screen latency that rises across the board, affecting agents on different networks and different computers at the same time. When that happens, the bottleneck is not the agent's connection — it is the server.

What agent screen latency actually measures

The VICIdial agent screen sends a small keep-alive request to the web server at regular intervals. Latency, as logged by the system, is the round-trip time for that request: how long it took from the browser sending the packet to receiving the response. This is not voice audio latency — it is HTTP response time. But it reflects the same underlying resource: the server's ability to answer requests quickly.

Under light load, the web server answers those keep-alives in a few dozen milliseconds. Under heavy load — many Concurrent calls, a busy database, or high CPU from Asterisk processing — the web server has to queue incoming requests. The round-trip time climbs accordingly.

The relationship between server load and latency

A VICIdial server handles several workloads simultaneously: Asterisk processes audio for every active Channel, the database records every status change and keep-alive, and the web tier serves all the agent and admin screens. When overall system load rises, the database query that backs the keep-alive response takes longer, which pushes up the measured latency.

This means a shift-wide latency rise is a proxy for overall server stress. If you chart it against system load and channel count, you will often see them track together. Latency starts climbing a few minutes before the system load number reaches concerning levels, which makes it a leading indicator rather than a lagging one.

flowchart LR
  A["More concurrent calls"] --> B["Higher channel count"]
  B --> C["Asterisk CPU rises"]
  C --> D["DB query queue grows"]
  D --> E["Keep-alive response slows"]
  E --> F["Latency climbs for all agents"]
  F --> G{"Isolated agent?"}
  G -->|Yes| H["Network or browser issue"]
  G -->|No| I["Server capacity problem"]

How to use latency data as a capacity signal

The Agent Latency Report gives you per-agent latency readings across a day. To use it as a capacity signal, look at the shift-wide average rather than individual agent spikes. A healthy system holds steady at a low baseline throughout the day. If the average climbs significantly during peak dialing hours and then drops once the dial level comes down, the load curve is driving latency.

One useful check is to compare the latency numbers on days where you ran a higher Dial level against days where you ran lower. If latency is noticeably higher on heavy-dial days, the server is showing you its ceiling. The next step is either to reduce the dial ratio at peak times or to move to a larger box.

When latency is not a server problem

Not every latency spike is a capacity problem. An individual Agent with consistently high latency while everyone else is fine is almost certainly a local network or browser issue. VPN tunnels with high overhead, browser tabs consuming CPU, and congested home routers all show up as high latency for one person only.

Latency gaps — periods where the log goes silent for an agent who should be logged in — are a different issue entirely and indicate that the agent's connection to the server was interrupted. That is covered in a separate post.

Putting it together with other health data

Latency is one signal among several. For a complete picture of how your server is handling load, pair it with system load, CPU breakdown, and channel count from the Server Performance Report — all part of the broader server health and capacity monitoring workflow. For a deeper look at what system load numbers mean on their own, see VICIdial system load explained.

If you are regularly hitting capacity limits, it may be time to look at what a larger managed box costs — see VICIfast plans and get a right-sized server live in under 40 seconds.