What breaks first when VICIdial runs at high volume

When VICIdial runs at high volume, the first thing to break is usually not the software but a limit around it: carrier channels, CPU at the busiest hour, the database, recording disk or the lead lists themselves. A Predictive dialing floor at three lines per agent puts about 150 live calls on the server at 50 agents and 300 at 100 agents. Each of those limits has a warning sign you can watch before agents feel it.

1. Carrier channels and calls per second

The first wall is often the trunk. If your carrier allows fewer Concurrent calls than the dialer wants to place, calls fail with busy or CONGESTION codes and agents wait. Ask for channels above your peak live calls (roughly 180 channels for 50 predictive agents), and check the calls-per-second limit too, since a predictive dialer places calls in bursts.

Warning sign: a rise in failed calls at the top of each hour, when every agent comes back from break at once. Our post on carrier concurrent channels shows how to read it, and the server health and capacity guide covers the rest of the monitoring.

2. CPU at the busiest hour

VICIdial's load follows live calls, not logged-in agents. Recording, Codec translation and AMD (answering machine detection) all add CPU per call. A server that is fine at 10am can lag at the noon peak.

Warning sign: agents report lag, or the screen shows a short freeze before a call connects. Read agent latency as a capacity signal for how to spot it early.

3. The database

Every call writes several rows, and the call logs grow every day. Large reports run against the same database the dialer uses, so a supervisor pulling a 90-day report at 11am can slow the floor.

Warning sign: reports that took seconds now take minutes, and the dialer stalls while they run. At cluster size, reports should read from a replica.

4. Recording disk

With every call recorded, a 50-agent floor writes about 15 GB of audio in an 8-hour day, and a 100-agent floor about 30 GB. A full disk stops recordings and can stop the dialer.

Warning sign: free disk falling by more than a day's worth each day. Move older recordings off the server on a schedule.

5. Lead lists

High volume burns through lists fast. When the Hopper runs dry, agents sit idle with the dialer running.

Warning sign: a falling count of dialable leads. Plan Lead recycling rules and load new lists before you need them.

6. The drop rate

As the dial level rises to keep many agents busy, dropped calls rise too. Keep the Drop rate under 3% per campaign over 30 days, and let the adaptive dial methods hold it there.

flowchart TD
  A[More agents] --> B[More live calls]
  B --> C{Carrier channels enough?}
  C -->|No| C1[Failed calls at peak]
  C -->|Yes| D{CPU headroom at peak?}
  D -->|No| D1[Agent lag]
  D -->|Yes| E{Reports on a replica?}
  E -->|No| E1[Dialer stalls during reports]
  E -->|Yes| F{Disk and lists planned?}
  F -->|No| F1[Recordings stop or hopper empties]
  F -->|Yes| G[Stable high-volume floor]

When one server stops being enough

One large server carries a lot. On VICIfast the biggest dedicated plan is sized for about 350 agents on a typical mix, and a 150-agent predictive floor fits on one dedicated server. We suggest a cluster from about 100 agents or 500 concurrent calls: a database with a replica for reports, web servers that also run the agents' phones, and dialer servers that only place calls. For the trade-offs, read single server vs clustered VICIdial.

To check your own numbers, use the agent capacity calculator. For what a high-volume setup looks like on VICIfast, see our VICIdial high-volume dialer page and the sizing page for 100+ agents. Server prices are on pricing.