Size your VICIdial cluster from the busiest minute, not the average
Your servers never run your daily average. They have to carry the busiest minute, because that is when calls drop. Here is how to find that minute and size from it.

Your dialer's numbers say the floor runs at a comfortable average, so a cluster sized for that average looks like enough. Then the busiest stretch of the shift arrives, every agent is logged in, and calls start dropping while agents sit ready.
The machines never run your average. They have to carry the busiest minute, because that's when the floor breaks.
What an average hides
An average spreads the quiet start and end of a shift across the busy middle. A floor that runs flat out for an hour can post a modest daily figure and still overload its servers at the same time every day.
Even careful measurements smooth things out. On 14 October 2026 we watched one outbound floor for a full shift: 166 Concurrent calls (calls up at the same moment) and 17 agents, averaged over 92 five-minute samples with calls up. A five-minute sample already flattens the single minute that hurt.
That's why our planning numbers carry room to spare. By our estimates one machine carries about 600 concurrent calls with room left over. Size against your peak and that spare room is still there when you need it. Size against the average and the peak eats it.
What the peak does to each machine
The dialers carry the calls. Asterisk, the phone engine on each dialer, does most of the work, and its load follows the calls that are up. Our estimate is about 2 cores of fixed load plus 0.03 cores per concurrent call, so the per-call part is what grows at the peak.
Dial harder and the calls-up count rises even when the headcount doesn't. One floor ran 18 agents with the Auto dial level (how many lines VICIdial dials for each waiting agent) between 4 and 8, on 4 cores. It showed 25% CPU pressure, work waiting its turn for a core. On 8 cores, that fell to 6%.
The database feels the peak differently. We replayed a live floor's real database traffic at five times its speed, about 900 calls. The database averaged 1.6 cores and peaked at 2.5, so the processor was fine. The slowest queries weren't.
At about 540 calls, the slowest 1 in 1,000 queries took 24 ms. At about 900 calls they took 208 ms, while 99% of queries still finished in under a millisecond. The typical query looked healthy; the worst moments got almost nine times slower. That tail came from table locks on VICIdial's call log, and the full replay results show how it grew.
Why the busiest minute is when calls drop
VICIdial hands each answered call to a free agent through the database. When the tables it needs are locked, the hand-off waits, but the drop timer keeps counting. Wait long enough and the customer is dropped, and your Drop rate climbs while agents sit ready.
There are two usual causes: pacing that dials more than the floor can take, and a locked database. Both get worse as the floor gets busier, which puts them in the same few minutes of your day.
flowchart TD
A[Daily average looks fine] -.-> B[Busiest minute: most calls up]
B --> C[Dialers busiest]
B --> D[Call log tables locked longer]
C --> E[Answered call waits for an agent hand-off]
D --> E
E --> F[Drop timer keeps counting]
F --> G[Customer dropped while agents sit ready]Find your busiest minute, then size from it
Start with the most calls you had up at once on your busiest day. Our post on what "most concurrent calls" tells you shows where to read it.
Count agents at that same peak. We build one dialer per 300 concurrent calls or 25 agents, whichever needs more, and never fewer than two. A peak of 600 calls and 50 agents takes two dialers. If the real busiest minute reaches 900 calls and 75 agents, it takes three, and a cluster sized from the lower number is one dialer short.
Then add the things that change the layout: the dial level you run when the floor is full, inbound and recording. With your agent count, those are the four things we ask on a sizing call. How many dialers a cluster needs covers the count in detail, and our guide to VICIdial clusters walks through the rest of the layout.
Frequently asked
- The shorter the window, the closer you get to what the machines actually carry. An hourly or five-minute figure is an average too, so it sits at or below the real peak. Leave room to spare above it.
- Yes. We size dialers on both: one per 300 concurrent calls or 25 agents, whichever needs more, and never fewer than two.
› Should I size from the busiest hour or the busiest minute?
› If I know my peak calls, does my agent count still matter?
If your busiest minute still fits on one server, start on a plan. If it doesn't, bring your peak numbers and book a sizing call to see today's cluster prices.
Related from VICIfast
About VICIfast
We run VICIdial servers for call centres, so you don’t have to. You get the dialer set up, secured and kept running — we handle the server, the updates and the backups. Bring your own phone carrier at no markup, or use VICIfast Voice, our own US calling and numbers, from $0.0085 a minute. From $49 a month per server.
Questions? Call +1 636 556 0022 and talk to someone who runs dialers.
Citing this article
VICIfast Engineering. “Size your VICIdial cluster from the busiest minute, not the average”. VICIfast LLC, November 20, 2026. Retrieved from https://vicifast.com/blog/size-vicidial-cluster-busiest-minute
Have questions?
Related posts
You might be interested in
Guides & tutorials
When a CRM polls VICIdial once per agent: the load you don't see
Guides & tutorials
Idle campaigns left active: the hidden database load in VICIdial
Guides & tutorials
Why searching a lead by phone number freezes VICIdial, and the missing index

Guides & tutorials
Keep the call log small: log archiving on a busy VICIdial database
VICIfast newsletter
Liked this? Get the next one in your inbox.
We ship the kind of stuff you just read - concrete, numbers-first, no drip. One email when a new post goes live. Unsubscribe in one click.
Comments
No comments yet - be the first.


