Asterisk performance and load-test planning
A server's call count is meaningful only with its workload and acceptance criteria. Describe call arrival rate, simultaneous calls, duration, codecs, media handling, recording and application dependencies before comparing hardware or changing settings.
Source-reviewed, 19 October 2026. Prepared from the cited documentation. Phone Guides has not independently load-tested this configuration. Scope: measurement planning; no server sizing guarantee or benchmark is supplied.
Describe the workload
| Measurement | Why it belongs in the result |
|---|---|
| New call attempts per second | Separates setup pressure from ongoing conversations |
| Simultaneous answered calls | Describes sustained call work |
| Codec and media path | Identifies transcoding and whether media crosses Asterisk |
| Calling features | Includes conferences, recording, queues and speech services |
| Registrations/subscriptions | Includes account churn and state notifications |
| Database/service requests | Exposes external dependencies and cache behaviour |
| Log/recording writes | Describes storage and diagnostic activity |
Record the exact Asterisk build, modules, operating system, hardware/VM allocation and network topology. A result from a different codec mix or direct-media arrangement may not predict yours.
Read tuning guidance as conditional
The official performance guidance discusses PJSIP and database/cache choices. Apply a change only after identifying the observed constraint and its effect on behaviour. Timer values and identification order should not be copied across deployments without their conditions.
Use the system requirements to identify upstream requirements, not as a calls-per-server table. Check codecs and transcoding before attributing a resource difference to server size alone.
For a PJSIP task queue investigation, record the exact Asterisk minor release before following the taskpool section of the performance guide. PJSIP Taskpool support starts in Asterisk 22.8.0; earlier 22.x releases follow the earlier threadpool guidance. On a supported release, core show taskprocessors can show task counts for the live pools. Compare the relevant PJSIP pool during the same repeatable traffic window. Counters alone are not a sizing result; confirm that call setup or another agreed service measure degrades at the same time before adjusting pool settings.
Measure the service, not just the process
Collect call setup/answer outcomes, audio symptoms, application response time and resource use over the same time window. Include CPU, memory, disk activity, network loss and relevant queue/backlog measurements. Preserve the collection method and sampling interval.
Define an acceptable failure rate and response-time range before a run. A process remaining alive while calls fail does not meet a service target. Check the traffic generator's own limits so a plateau is not mistaken for the PBX limit.
Change one explanation at a time
Establish a baseline, make one justified change, then repeat the same workload. Preserve the previous configuration and restore it if the change harms call handling. Increasing worker limits can move pressure to another resource or dependency.
Do not apply load to a public carrier, production phones or another organisation's service without the intended test arrangement. A local scenario can use controlled destinations and dependencies; its results remain scoped to those conditions.
Reader checks
When testing becomes available, include steady traffic, a short burst, long calls, recording/storage pressure and a slow external dependency. Inspect recovery after traffic drops. Record completed calls and errors alongside resource charts, not only the largest attempted count.
Use monitoring and alerts for operational observation. Keep an untested sizing estimate labelled as an estimate until measured against its workload and acceptance criteria.
Return to Asterisk documentation.