Signs your VICIdial server needs more capacity

A VICIdial server does not fail suddenly. It sends warning signs for hours or days before calls start dropping or agents freeze in their screens. Knowing how to read those signs means you can act before your dialing shift turns into a support emergency. This post covers the concrete indicators that tell you a box is nearing its ceiling.

System load that no longer comes back down

Linux system load is a run-queue average: how many processes wanted CPU time but had to wait. On a box with eight cores, a load of 8 means every core is busy but nothing is queuing. A load of 16 on that same box means half of the work in line is waiting. A healthy dialer climbs during peak dialing and drops during quiet hours. The problem sign is not a high peak — it is when load stops dropping between shifts. If your overnight load is still sitting at the same level as your afternoon peak, the box is not catching up. You can see both average load and peak load for any shift inside the Server Performance Report.

Channel count that bumps against a ceiling

Every leg of every call — the outbound dial, the Agent answer, a transfer — opens a Channel on the Asterisk process. A VPS that was sized for 50 Concurrent calls starts to struggle when you run 70 without upgrading. The symptom is subtle at first: audio quality slips, calls that should be answered ring a moment too long, and agents report short one-way-audio events at the start of a live connection. The channel count graph from the Server Performance Report will show a flat top on your busiest hours if you are hitting the ceiling — the system is not routing additional calls, it is just capping out.

IDLE CPU collapsing during peak hours

The processor time breakdown splits into USER (Asterisk, PHP, Perl application work), SYSTEM (kernel, network, disk interrupts), and IDLE (spare headroom). A well-sized box running a predictive campaign usually keeps IDLE somewhere above 20 percent even during busy periods. When IDLE collapses to single digits for sustained stretches, the CPU has no slack left for burst work like answering an inbound call or running a dispositioned lead through a routing script. If SYSTEM is also high at the same time, you may have disk or network saturation stacking on top of call processing pressure.

How the warning signs connect

flowchart TD
  A["Busy shift starts"] --> B{"Load stays high after shift?"}
  B -->|No| C["Normal — box is healthy"]
  B -->|Yes| D{"IDLE CPU near zero?"}
  D -->|No| E["Check for runaway process"]
  D -->|Yes| F{"Channel count flat-topped?"}
  F -->|No| G["DB or disk pressure — investigate"]
  F -->|Yes| H["Capacity ceiling reached — resize needed"]

These three signals rarely appear in isolation. Sustained load, flat-topped channel graphs, and collapsed IDLE tend to arrive together because they are measuring the same underlying constraint from different angles. When you see all three on the same shift, the box has reached its practical ceiling for that call volume.

Other signals worth watching

• Disk usage rising steadily without cleanup — Call recording files and log rotation both consume disk; a full disk will stop recordings and can stall the database.
• Keepalive processes restarting frequently — the keepalive scripts restart Asterisk sub-processes when they go missing; repeated restarts mean the box cannot sustain those processes under load.
• Agent screen delays and disposition timeouts — when the web server and database are competing with Asterisk for CPU, agent-facing pages slow down noticeably.
• Increasing answer seizure ratio failure rate — the dialer places a call, Asterisk answers but cannot bridge the leg fast enough, and the call drops before it reaches an agent.

For a deeper look at what each metric means and how to set up tracking across shifts, see the server health and capacity guide. For context on what system load numbers actually represent, read VICIdial system load explained.

On a Single tenant managed host, resizing is a straightforward operation rather than a migration project. If you want a box that is already sized for your call volume and monitored from first boot, take a look at VICIfast plans.