The same kernel grows into completely different shapes across six scenarios. Understanding the shapes is more useful than memorising configuration parameters.
Asterisk documentation is full of configuration options, but very little of it tells you what people actually build with it. Here are the six most common deployment shapes, each with its own centre of gravity.
The classic pattern: one server, anywhere from dozens to hundreds of SIP extensions, one or two external trunks. The requirements are stability, maintainability and mobile support. Most deployments put a graphical layer such as FreePBX on top so configuration moves from the command line to a browser.
Operations focus here is backup and permissions. Extension configs, recordings and voicemail are business data, and losing them is painful.
The focus shifts from calls to queuing. Queue strategy, agent states, overflow rules, recording quality checks and wallboards decide whether the system can carry the business. Asterisk's Queue application covers most of the baseline; complex routing and quality analysis usually need ARI code.
Voice menus look simple and are the hardest thing to design well: keep the tree shallow, keep the option count low, always offer a route to a human. Technically it is a dialplan plus audio files; the real work is process design and recording production.
Internet telephony providers use Asterisk as access and application layers, characterised by high concurrency, mixed protocols and strict billing. These deployments demand the most performance tuning and protocol interoperability work, down to kernel parameters, NIC interrupts and media transcoding paths.
Hardware vendors embed Asterisk in their own products to build intercom systems, emergency call pillars and dispatch consoles. Endpoints are unusual, networks are poor and offline operation is mandatory. Many designs trim Asterisk to a minimum, keeping only the channel drivers they need.
In embedded scenarios, how much you can delete matters more than how much you can add.
Automated outbound calls, voice OTPs and alarm announcements. Technically the simplest, with two engineering traps: carrier limits on volume and content, and the handling of retries and delivery receipts. Many projects discover after launch that the real cost is knowing whether the call actually connected.
Whatever the shape, three issues always appear: NAT and firewalls, recording storage, and high availability. The first two are deployment problems; the third is architectural. HA ranges from active-standby with shared storage to a front-end SIP proxy doing load distribution, and the investment varies enormously — choose according to how much downtime the business can tolerate.