We pushed a VICIdial database to 7 times a live floor: what slowed first
We replayed a real floor's database traffic at up to seven times its speed. The processor barely noticed. The slowest queries told a different story.
How far can one VICIdial database go before your floor feels it? And when it finally slows, is it the processor running out, or something else?
Rather than guess, we took 5 minutes of a live outbound floor's real database traffic and played it back, faster and faster, on one machine. The answer pointed away from the processor.
How the test worked
The traffic came from a busy outbound floor we measured on 14 October 2026, running 166 Concurrent calls (calls up at the same moment) on one server. At normal speed our 5-minute slice ran 2,155 queries a second, about what 180 calls produce.
On 15 October we replayed that slice at 1x, 3x, 5x and 7x on a 16-core AMD Ryzen 9 7950X3D with 128 GB of memory and mirrored NVMe disks. It's the same kind of machine we use for a cluster's database. Seven times the floor works out to roughly 1,260 calls.
The results
| Speed | About calls | Queries/s | DB cores avg / peak | 99% of queries | 99.9% of queries |
| 1x | 180 | 2,155 | 0.4 / 1.3 | 0.45 ms | 18 ms |
| 3x | 540 | 6,464 | 1.1 / 2.1 | 0.6 ms | 24 ms |
| 5x | 900 | 10,774 | 1.6 / 2.5 | 0.69 ms | 208 ms |
| 7x | 1,260 | 15,083 | 2.1 / 3.1 | 0.77 ms | 290 ms |
Read the table in two halves. The processor columns hardly moved: from 0.4 cores on average at 1x to 2.1 at 7x. The database never needed more than 3.1 of its 16 cores.
The 99% column hardly moved either. Even at 7x, 99 queries in 100 finished in under a millisecond.
The last column is where it changed. The slowest one query in a thousand took 18 ms at normal speed and 24 ms at 3x. At 5x it jumped to 208 ms, and at 7x to 290 ms.
What slowed first: locks, not cores
That slow tail came from table locks on VICIdial's call log. VICIdial's call records live in MyISAM tables, and a MyISAM table is locked as a whole while any read on it runs. Anything that needs to write to it waits until the read finishes, no matter how many cores sit idle.
flowchart LR
R[Report reads the call log] --> L[Whole table locked]
L --> D[Disposition save waits]
L --> H[Call hand-off waits]
D --> S[Agent screen hangs]
H --> T[Drop timer keeps counting]Why care about one query in a thousand? Volume. At 7x the database answered 15,083 queries a second, so one in a thousand is about 15 slow queries every second.
Some of those are a Disposition (the status an agent picks when a call ends) being saved. Others are an answered call being handed to a free agent, which VICIdial does through the database. If the table is locked, the hand-off waits but the drop timer keeps counting, and a call that waits too long lands in your Drop rate (dropped calls as a share of calls a person answered).
That's why a bigger database machine isn't the fix at this size. The queries in the slow tail aren't short of processor. They're queuing.
What we do about it
The fix is keeping the call log small with VICIdial's own log archiving. VICIdial ships a script for it, ADMIN_archive_log_tables.pl, which moves old rows out of the log tables. Run it on a schedule.
Then take what reads the call log during the shift off the floor's path: reports, stats tools, Lead searches (looking up a customer record). On one floor, moving a single stats tool off the call log cut lock waits from 58 a second to 13.
Not sure which of your reports are the heavy ones? Our guide to the Admin Report Log Viewer shows how to spot slow-running reports, and monitoring database load covers what to watch day to day.
For where the database sits among the other machines, see the full cluster guide. If your floor fits on one server, our plans are where to start. If you're planning for hundreds of calls, see today's cluster prices and book a sizing call.
Related from VICIfast
- How we build VICIdial clustersA database, dialers and a standby, sized from load we measured on a live floor.
- Bare metal VICIdial serversEvery core yours, mirrored NVMe, delivered in minutes to days. Priced from live stock.
- Bring your own server - $29/moYour VPS or on-prem hardware; we install and manage VICIdial over SSH.
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. “We pushed a VICIdial database to 7 times a live floor: what slowed first”. VICIfast LLC, November 19, 2026. Retrieved from https://vicifast.com/blog/vicidial-database-replay-7x-load
Have questions?
Related posts
You might be interested in
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.



