docs: explicitly explain how to use without worker mode (#1275)

Author: fellmann

README.md

@@ -60,7 +60,8 @@ frankenphp php-cli /path/to/your/script.php
 
 ## Docs
 
-* [The worker mode](https://frankenphp.dev/docs/worker/)
+* [Classic mode](https://frankenphp.dev/docs/classic/)
+* [Worker mode](https://frankenphp.dev/docs/worker/)
 * [Early Hints support (103 HTTP status code)](https://frankenphp.dev/docs/early-hints/)
 * [Real-time](https://frankenphp.dev/docs/mercure/)
 * [Configuration](https://frankenphp.dev/docs/config/)

docs/classic.md

@@ -0,0 +1,9 @@
+# Using Classic Mode
+
+Without any additional configuration, FrankenPHP operates in classic mode. In this mode, FrankenPHP functions like a traditional PHP server, directly serving PHP files. This makes it a seamless drop-in replacement for PHP-FPM or Apache with mod_php.
+
+Similar to Caddy, FrankenPHP accepts an unlimited number of connections and uses a [fixed number of threads](config.md#caddyfile-config) to serve them. The number of accepted and queued connections is limited only by the available system resources. The PHP thread pool operates with a fixed number of threads initialized at startup, comparable to the static mode of PHP-FPM.
+
+Queued connections will wait indefinitely until a PHP thread is available to serve them. To prevent that, set a reasonable [write timeout in Caddy](https://caddyserver.com/docs/caddyfile/options#timeouts).
+
+Each Caddy instance will only spin up one FrankenPHP thread pool, which will be shared across all `php_server` blocks.

docs/performance.md

@@ -8,7 +8,7 @@ However, it is possible to substantially improve performance using an appropriat
 By default, FrankenPHP starts 2 times more threads and workers (in worker mode) than the available numbers of CPU.
 
 The appropriate values depend heavily on how your application is written, what it does and your hardware.
-We strongly recommend changing these values.
+We strongly recommend changing these values. For best system stability, it is recommended to have `num_threads` x `memory_limit` < `available_memory`.
 
 To find the right values, it's best to run load tests simulating real traffic.
 [k6](https://k6.io) and [Gatling](https://gatling.io) are good tools for this.