Scaling a cloud VICIdial box up for more agents

VICIdial's resource requirements grow in two directions at once: CPU climbs with Concurrent calls and transcoding load, while RAM grows with the number of active SIP channels and database connections. When your cloud VPS starts falling behind, the fix is usually a vertical resize — moving to a larger instance type with more vCPU and RAM on the same server.

This post covers how to plan capacity before you hit the wall, what a resize actually involves, and the signals to watch for that tell you it is time to scale up.

Where VICIdial spends CPU and RAM

Asterisk is the heaviest CPU consumer. Each active audio channel requires real-time codec processing, and if you are doing Recording format (WAV/MP3) conversion on the fly (writing WAV then transcoding to GSM), that adds another encoding pass per call. As a rough rule, plan for about one vCPU per 10 to 15 Concurrent calls under normal load. This assumes G.711 (no transcoding) — if you are bridging between G.711 codec and G.729 codec, the CPU cost per call is higher because the codec conversion is more intensive.

RAM is consumed by Asterisk channels and by the MySQL database. A working estimate is 50 MB per active SIP channel in Asterisk, plus another 512 MB to 1 GB for the database depending on your list sizes and active queries. Add a base of about 2 GB for the OS, VICIdial PHP processes, and other services, and you get a usable formula: required RAM = (max_concurrent_calls * 50 MB) + 2 GB base + DB overhead.

Mapping agent count to instance size

flowchart TD
  A[Agent count] --> B[lines-per-agent ratio]
  B --> C[Max concurrent calls]
  C --> D[CPU: 1 vCPU per 10-15 calls]
  C --> E[RAM: 50 MB per channel + 2 GB base]
  D --> F{Instance size}
  E --> F
  F -->|10-20 agents| G[4 vCPU / 8 GB]
  F -->|20-50 agents| H[8 vCPU / 16 GB]
  F -->|50-100 agents| I[16 vCPU / 32 GB]
  F -->|100+ agents| J[Consider dedicated bare-metal]

The Lines per agent ratio in your VICIdial campaign settings determines how many calls Asterisk is managing per logged-in agent. A predictive dialing campaign might run 3 to 4 lines per agent during aggressive dialing. That means 50 agents could generate up to 200 Concurrent calls at peak, which requires roughly 14 vCPUs just for Asterisk and around 12 GB of RAM for channels plus overhead.

The table below is a starting point, not a hard spec. Your actual profile depends on Dialer pacing, whether you are doing local recording, and how active your campaign lists are. Monitor CPU and RAM under real load before committing to an instance size.

Signals that you need to resize

Watch for sustained CPU steal above 5 percent (common on over-subscribed cloud hosts), load average consistently above vCPU count, Latency in the VICIdial web UI rising during peak hours, and Asterisk logging slow-frame warnings in the console. On the RAM side, watch for the box using swap — once VICIdial is swapping, call quality degrades and you will start seeing dropped channels and missed keepalive responses.

How a vertical resize works

On most cloud providers a vertical resize requires a brief shutdown of the VPS. The provider migrates the disk to a larger instance type and boots it back up. This typically takes 3 to 10 minutes. VICIdial will come back up cleanly if Asterisk and MySQL are set to start on boot, which they are by default after a standard install. Active calls will drop during the downtime, so schedule the resize at the end of a dialing shift or during a quiet weekend window.

Before resizing, take a Server snapshot so you have a rollback point. After the resize, verify that all services came back up with systemctl status asterisk mysql vicidial and run a test call before agents log in. For more on disk sizing as you scale up, see sizing cloud disk for call recordings.

VICIfast and managed resizing

On VICIfast Managed hosting, resize operations are handled through the dashboard without SSH access. The platform coordinates the shutdown, resize, and boot sequence and notifies you when the box is back online. [[Provisioning]] of a new box from scratch takes under 40 seconds if you ever need to move to a fresh instance. The full background on running VICIdial in the cloud is in the VICIdial cloud guide.

To see which instance sizes are available and what each plan includes, check the pricing page for current options.