Network buffers on a busy VICIdial dialer: where voice packets get lost
Choppy audio isn't always the network. On a busy VICIdial dialer, desktop-sized Linux buffers overflow and throw voice packets away on the server itself.
Agents say words cut out mid-sentence. Customers ask them to repeat themselves, and it's worst in the busiest hours. You check the carriers and the agents' connections, and both look clean.
Sometimes the voice packets aren't lost on the way at all. They reach the dialer and get thrown away there, before Asterisk ever reads them.
Where the packets wait
Call audio travels as RTP: a steady stream of small packets, each holding a slice of sound, flowing in both directions for as long as the call lasts. On a dialer, every customer call and every agent's phone adds streams of its own.
Packets don't arrive in a neat line. They come in clumps. Linux holds them in network buffers, small waiting areas in memory, until Asterisk picks them up. If a buffer is full when the next clump lands, Linux has nowhere to put it and discards it.
Nothing reports an error to the call. The sound simply has a hole in it, and to the people talking it's the same choppy audio as Packet loss on a bad connection.
flowchart LR
C[Carrier calls] --> NIC[Dialer network card]
A[Agent phones] --> NIC
NIC --> B[Linux network buffer]
B -->|room left| AST[Asterisk handles the audio]
B -->|buffer full| X[Packet thrown away]
X --> G[Gap in the call audio]Why the defaults overflow
Linux's default network buffers are sized for a desktop: a browser, a mail client, a download. A dialer is a very different machine. In our layouts one dialer carries up to 300 Concurrent calls, and each of them is sending and receiving audio the whole time.
On one 60-agent server we measured, default buffers threw away 458,000 voice packets in four days. The loss wasn't a steady trickle. It came in bursts, so a call could sound fine for minutes and then break up.
That's what makes it hard to pin down. Agents report it as the occasional bad call, the processor and memory graphs can look normal, and nobody thinks to look at the network inside the server.
Telling it apart from network trouble
- Network loss follows a route. Bad audio on one carrier, or for agents at one site or on one home connection, points outside the dialer. Our guide to choppy or robotic audio covers those causes, including Jitter.
- Buffer loss follows the server. It shows up across carriers and agents on the same dialer, while other dialers carrying similar calls sound fine.
- The server keeps count. Linux counts the packets it discards for lack of buffer room. If that count climbs during a shift, the loss is happening on the dialer.
The fix, and where it goes
Size the buffers for the job: room for hundreds of calls' worth of audio instead of a desktop's worth. Every server we build gets buffers sized that way before it carries its first call.
Bigger buffers don't rescue a dialer that is out of processor. If Asterisk can't keep up, packets still pile up and overflow. Then the answer is fewer calls per machine, which is the reason a dialer has a defined job and a limit in a cluster.
Our cluster guide shows how a cluster spreads calls across dialers. If one server carries your floor, pick a plan and we'll set it up the same way. If your floor needs more, book a sizing call and get a layout built for your call volume.
Frequently asked
- No. Most of the time the loss is on the network between the dialer and a carrier or an agent. Check whether the bad audio follows one route or the whole server before you change anything.
- Yes. Buffer sizing is set on each Linux machine, one box or many. A cluster helps when one machine is simply carrying more calls than it should.
› Is choppy audio always the dialer's fault?
› Can I fix this on a single server?
Related from VICIfast
- Running VICIdial across several serversA database, dialers and a standby, sized from load we measured on a live floor.
- Dedicated VICIdial hardware, live pricesEvery core yours, mirrored NVMe, delivered in minutes to days. Priced from live stock.
- See it running on a callA real server provisioned in front of you. No slide deck.
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. “Network buffers on a busy VICIdial dialer: where voice packets get lost”. VICIfast LLC, November 22, 2026. Retrieved from https://vicifast.com/blog/vicidial-dialer-network-buffers
Have questions?
Related posts

Guides & tutorials
Where agents connect in a VICIdial cluster: screen from one machine, phone from another

Guides & tutorials
How many dialer servers does a VICIdial cluster need?

Guides & tutorials
VICIdial clusters: how they work, when you need one, and how to size one

Guides & tutorials
What a dialer server does in a VICIdial cluster
You might be interested in
Guides & tutorials
Why customers get connected to silence when an agent's browser phone drops
Guides & tutorials
Why your dashboard's drop rate doesn't match VICIdial's report
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
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.