I first removed the Scanner and used the underlaying BufferedReader, but this broke the detection of line endings.

How so? BufferedReader.readLine() is documented to handle the 3 common EOL variants.

Also, out of curiosity, can you run a benchmark using the Scanner + InputStreamReader? For example:

this.scanner = new Scanner(new InputStreamReader(inputStream, encoding));

Scanner is going to do its own buffering, so maybe having the BufferedReader is adding unnecessary overhead.

Hi again,

@seanbright:
The Scanner exists, because Asterisk uses different line breaks at different places of manager events:

• normal lines are separated by \r\n and usually contain a key/value pair,
• in the CommandResponse (for a CommandAction), there is a special line containing the (multi-line) answer of the command separated by \n.
  To distinguish this correctly, lines have to be separated by \r\n only. So the fast BufferedReader can't be used here and this is why there is a Scanner.

Unfortunately, I can't post an example (as my browser runs in a restrictive sandbox). So I can only point you to ManagerReaderImpl, line 181: Special handling for "Response: Follows" (CommandResponse).

Futher, I did benchmarks over network. This went well for the Scanner, but the BufferedReader and LineReader were faster than the 1Gbps connection. The results looked something like this:

• with Scanner: ~32MB/s and 12% CPU (one core of the quadcore)
• without Scanner: ~52MB/s and 4% CPU (network limitation)

The Scanner without BufferedReader was a bit slower:
13958ms InputStreamReader(w/o BufferedReader)+Scanner

@bsutton:
In fact, our code use NIO socket (just some hacky replacement). But performance issues where the reason why we changed. The normal socket is missing some timeout features that were crucial for our application.
I think NIO sockets perform much better if you are dealing with lots of connections opening and closing and everything is running in parallel. This does not seam to be the typical use case for AMI connections.

@elara-leitstellentechnik given the above discussion do you still want to push this request?