Sizing a VICIdial server for your agent count

The first number people reach for when sizing a VICIdial box is agent count. It is the wrong number. What loads the server is concurrent calls, and a predictive campaign dials several lines per agent. Twenty agents at a 3-to-1 ratio is sixty live channels, each carrying audio the box has to mix, record, and bridge. Size for the calls, and the agents take care of themselves.

Start from concurrent calls

Your true load is concurrent channels. Multiply agents by your Lines per agent ratio to get peak Concurrent calls. A campaign on Predictive dialing at a 3-to-1 dial ratio with 30 agents peaks around 90 channels. Each of those is an RTP stream Asterisk processes in real time. The ratio, set by your Dialer pacing, is the multiplier that turns a modest agent count into real CPU work.

flowchart TD
  A[Agent count] --> C[Multiply by dial ratio]
  B[Lines per agent] --> C
  C --> D[Peak concurrent calls]
  D --> E[CPU cores]
  D --> F[RAM]
  D --> G[Disk for recordings]
  H[Recording on] --> G

Cores and RAM

Asterisk does the heavy lifting, and audio processing is CPU-bound. As a rough rule, modern cores handle several dozen concurrent G711 channels each, fewer if you transcode to G.729 codec, which is cheaper on bandwidth but costs CPU to encode. Pick your Codec with that tradeoff in mind. RAM is less of a bottleneck than CPU on a single-server build, but MySQL or MariaDB wants room for its buffers, and a busy hopper and reporting workload eat memory. Give the database real headroom rather than the minimum.

Two ratios to keep straight. Agents to calls is your dial ratio, which drives channels. Channels to cores is your hardware ratio, which drives the box. A bottleneck shows up as audio glitches and dropped calls long before the CPU graph pegs at 100 percent, so leave headroom above your expected peak. Sizing to your average load is a trap; campaigns spike, and a box sized for the average chokes on the peak. Size for the busiest hour of your busiest day, then add a margin on top of that.

Disk, and why it grows

If you record calls, disk is the line item that grows every day. A single channel of Call recording is a few megabytes per minute depending on Recording format (WAV/MP3), and at scale that is gigabytes a day. Size disk for your retention window, not for a single day. The database itself stays modest, but recordings on a high-volume box can dwarf everything else. Plan retention before you run out of space mid-shift.

One box or more

For most operations a single well-sized server runs the database, the web admin, and Asterisk together, and that is the simplest thing to operate. There is a point where you split roles across boxes, but it is higher than most people assume, and clustering adds operational cost. Our single server vs cluster post walks the decision, and the installation guide covers sizing during bring-up.

We run one customer per box, so your Single tenant server's resources are never shared with anyone else's calls, and the sizing math above is the whole picture. Need more headroom as you grow agents? Resize the VPS to a bigger tier instead of re-architecting. Provisioning a right-sized box takes under 40 seconds, and you keep root SSH to tune Asterisk for your ratio. See plans and pricing.