SIP Trunk CPS Limits for Predictive Dialers Explained
Calls per second caps how fast a dialer launches calls, not how many run at once. Work out the CPS a campaign needs and avoid 503 rejections at shift start.
Kurze Antwort
A SIP trunk CPS limit is the maximum number of new calls your carrier accepts per second, separate from how many calls can be active at once. To size it, divide the calls your dialer keeps ringing (agents times lines per agent) by the average seconds each attempt spends ringing and in answering-machine detection. Then allow for the burst at campaign start, where most 503 rejections happen.
Every SIP trunk has two ceilings that people mix up. One is how many calls can be active at the same moment, the channel count. The other is how quickly new calls can be started, calls per second or CPS. A predictive dialer stresses both, but a campaign can sit comfortably below its channel limit and still see a wall of failed calls in the first seconds of a shift because it exceeded the CPS limit. This guide explains the difference, shows how to estimate the CPS a campaign needs with simple arithmetic, and covers how to handle the limit in dialer settings and in the conversation with your carrier.
All figures in the worked examples are illustrative inputs, not benchmarks or promises. Your own ring times, dial ratios and carrier limits will differ.
CPS versus concurrent channels
Think of a trunk as a door. Channels are how many people can be inside the room at once. CPS is how many people can walk through the door each second. A room with 200 places can still jam if 150 people try to enter in the same second.
| Channels (concurrent calls) | CPS (calls per second) | |
|---|---|---|
| What it limits | Calls active at the same moment | The rate at which new calls start |
| What drives it | Agents, dial ratio, call length | Pacing, ring time, campaign starts, retries |
| When it bites | At the busiest sustained period | In bursts: start of shift, after breaks, after an outage |
| Typical symptom | Calls fail when all channels are busy | A burst of rejections that clears when the rate drops |
Carriers define CPS in their own way. Some measure it strictly per second, others over a short window, and some apply it per account, per trunk or per source IP. Ask which, because the answer changes how you pace (see the carrier conversation below).
Why predictive campaigns burst
A predictive dialer places more calls than there are free agents, because many calls will not be answered. At steady state it launches calls continuously as attempts end. The bursts come from specific moments:
- Campaign start. At the first second, no attempt has ended yet, so the dialer wants to bring its full set of in-flight attempts up. Without pacing, that is the whole set launched together.
- Mass login. Many agents becoming available at once, for example at the start of a shift, creates a spike of demand for lines.
- After breaks and lunch. When a team returns together, the dialer catches up.
- After a failure. If a trunk drops and recovers, queued retries and the normal flow stack up together.
- Several campaigns on one trunk. Each can be inside the limit while their sum is not.
- Retries of rejected calls. A rejection, retried at once, adds load to the very limit that caused it.
The effect is easy to misread. A shift-start burst of failures looks like a routing or number-quality problem, but it ends within minutes and returns the next morning. The pattern, rather than any single error, is the clue. If you see the same failure only at the top of the hour, think rate.
Working out peak CPS
You do not need a simulation to get a usable estimate. In steady state, the launch rate equals the number of attempts in the setup stage divided by the average time an attempt stays in that stage. This is the same relation that governs any queue (Little's law, applied loosely here). In dialer terms:
Steady-state CPS ≈ (agents × lines per agent) ÷ (average ring time + AMD time, in seconds)
"Lines per agent" is the dial ratio, the number of simultaneous attempts the dialer keeps going for each available agent. "Ring time plus AMD time" is how long an attempt occupies the setup stage before it is answered and handed over, or ends. A call that is never answered occupies the stage for up to the ring timeout, so use the average, not the maximum.
Praxisbeispiel (zur Veranschaulichung)
Suppose 40 agents are available, the dialer keeps 3 lines per agent, and attempts spend on average 25 seconds in ringing and answering-machine detection.
- Attempts in setup: 40 × 3 = 120.
- Steady-state CPS: 120 ÷ 25 = 4.8 calls per second.
Now the burst. If the dialer launches all 120 attempts the moment the campaign starts, it asks for 120 new calls within the first second or two: 60 to 120 calls per second, roughly 12 to 25 times the steady-state rate. The steady figure is the sustained need; the start-up figure is the peak that trips limits.
The same team with different ring settings gives a different answer. This table uses the same 120 attempts in setup:
| Average time in setup | Steady-state CPS (120 attempts) | Reading |
|---|---|---|
| 15 seconds | 8.0 | Short ring timeout: more calls per minute, a higher rate. |
| 20 Sekunden | 6.0 | |
| 25 seconds | 4.8 | The example above. |
| 30 Sekunden | 4.0 | Long ring timeout: fewer launches, lower rate. |
Two lessons follow. Shorter ring timeouts and faster detection increase the launch rate for the same agents, so a tuning change that improves throughput can quietly push you over a CPS limit that was previously fine. And a team doubling its agents doubles the steady rate. Add the rates of every campaign sharing the trunk, and keep a margin: a common practice is to plan well under the limit rather than at it, since real ring times vary and bursts are never perfectly smooth.
Channels are a separate calculation. Your concurrent calls are the attempts in setup plus the calls in conversation. For the example above, 120 in setup plus agents on live calls means the trunk needs comfortably more than 120 channels, which is why the BYOC setup guide suggests sizing for peak plus headroom.
What happens when you go over the limit
Carriers enforce CPS in different ways. Some queue the excess briefly, some drop it silently, and others reject the extra calls with a SIP error, commonly 503 Service Unavailable. RFC 3261 itself allows a server to refuse the connection or drop the request instead of answering 503, so silence can be the symptom too. It defines 503 as the server being temporarily unable to process the request because of overload or maintenance, and says the server may include a Retry-After header telling the client when to try again. A client receiving it should try an alternate server and should not send further requests to that server for the Retry-After period, if one is given. If there is no Retry-After, the client treats the response as a 500. Our SIP error codes guide explains the other codes you may see alongside it.
The cost to a campaign has three parts:
- Lost attempts. Each rejected call is a contact you did not reach. If the dialer records it as no answer, the contact also loses an attempt from its retry allowance.
- Wasted agent time. A predictive dialer that expects a number of answers from its launches finds fewer, leaving agents idle until the next round.
- Distorted data. Contact rate and answer rate look worse than the list deserves. That can send tuning in the wrong direction. Before changing the list or the script, check whether the rejections cluster at the start of the hour. See the contact rate guide for the other levers.
Pacing settings and staggered starts
The fix is nearly always on the sending side: make the dialer launch at a rate the trunk accepts.
- Set the dialer CPS limit at or below the carrier limit. Dialer gateways commonly have such a field. In the DialerBee gateway form the CPS limit sits on the Routing tab and ships at 20; set it from your carrier's actual number.
- Ramp, don't jump. Start campaigns with a short ramp, adding lines per agent in steps rather than all at once.
- Stagger campaign starts. If five campaigns begin on the hour, move them a minute or two apart.
- Stagger agent logins where you can, or have the team start in two waves.
- Lower the dial ratio temporarily, using the supervisor guardrails for minimum and maximum lines per agent that the Vorhersage-Wählgerät provides.
- Delay retries of rejected calls by a few seconds and honour any Retry-After value instead of retrying at once.
- Split traffic across carriers. Two trunks with their own limits add their rates together; the Priority and Weight fields in the gateway form exist for this.
Der Leitfaden zur Optimierung des Predictive Dialers covers ring time, AMD and abandon-rate settings, which are the other half of the equation.
Talking to your carrier
Rate limits are a commercial and technical setting at your carrier, and they can usually be discussed. Before the call, work out your peak and steady figures as above. Then ask:
- What exactly is the CPS limit, and is it per account, per trunk or per source IP?
- Over what window is it measured: a strict second, or an average over several seconds?
- What happens above the limit: queued, dropped or rejected, and with which SIP code and Retry-After value?
- Can the limit be raised, and is there a charge or a different plan for it?
- Is there a separate burst allowance, for example for campaign starts?
- Does the channel limit count calls in the ringing stage as well as answered calls?
- Do outbound retries from your side count against the same limit?
Get the answers in writing and put the numbers in your gateway configuration. A limit nobody has written down is the one that causes the incident.
Wie DialerBee passt
DialerBee supports bring-your-own-carrier trunks; the BYOC-Funktionsseite describes carrier connection, multiple trunks and failover, and the setup guide has a routing and capacity step with channel and CPS fields on the Routing tab, plus advice to test a small burst before going live. The Vorhersage-Wählgerät paces calls against agent availability with supervisor guardrails on lines per agent. For background on what the dialer is doing, see what a predictive dialer is.
Checklist before the next campaign
- Carrier CPS, measurement window and above-limit behaviour written down.
- Steady-state and start-up CPS estimated for every campaign on the trunk.
- Dialer CPS limit set at or below the carrier limit, with margin.
- Campaign starts and agent logins staggered.
- Retry of rejected calls delayed, and failed records requeued rather than lost.
- A burst test run before the first live campaign.
Häufig gestellte Fragen
What is a SIP trunk CPS limit?
It is the maximum number of new calls per second your carrier accepts on the trunk. It is separate from the channel count, which limits calls active at the same time.
How is CPS different from concurrent channels?
Channels limit how many calls are in progress at once; CPS limits how fast new calls can start. A trunk can have plenty of free channels and still reject calls if too many are started in one second.
How do I calculate the calls per second a predictive dialer needs?
Multiply the number of available agents by lines per agent, then divide by the average seconds an attempt spends ringing and in answering-machine detection. For 40 agents, 3 lines and 25 seconds, that is 120 divided by 25, or 4.8 CPS in steady state. Plan separately for the start-of-campaign burst.
Why do I get 503 errors at the start of a campaign?
The dialer launches many calls at once, and the carrier rejects those above its rate limit, often with 503 Service Unavailable. It clears as the launch rate falls. Lower the dialer CPS limit, ramp up lines per agent and stagger campaign starts.
Should I set the dialer CPS limit equal to the carrier limit?
Set it at or below the carrier limit, with some margin, because real ring times and bursts vary. Also add together all campaigns that share the trunk.
Does a higher CPS limit improve contact rate?
Not by itself. It removes rate-related rejections, which stops you losing attempts, but contact rate still depends on list quality, calling windows, caller ID and pacing.
Quellen
- RFC 3261: SIP, section 21.5.4 (503 Service Unavailable, Retry-After, and dropping instead of responding) (RFC Editor, checked November 7, 2026)
Note: The worked examples use illustrative numbers to show the arithmetic. They are not benchmarks, and the estimate is an approximation. Carrier definitions of CPS and enforcement vary, so confirm them with your carrier.
Weiterführende Lektüre
Verwandte Artikel
Bereit, DialerBee in Aktion zu sehen?
Fordern Sie eine 15-minütige Live-Demo an oder starten Sie eine 14-tägige kostenlose Testphase – ohne Folien, ohne Verpflichtung. Das Telefonieren beginnt, sobald Ihr eigener SIP-Trunk oder Ihre TK-Anlage verbunden und freigegeben ist.
14-tägige kostenlose Testphase · kostenloses KI-Guthaben · keine Kreditkarte · 11 Sprachen · BYOC · Kontrollen zur Unterstützung der Compliance