13.10 is better by the simple fact 2.1.x branch is officially deprecated and abandoned.
This means two simple things:
- No new version in the 2.1.x will be released.
- Bugs found in 2.1.13 might not get fixed.
Also I would suggest not profiling but rather measuring the real-life performance using a suitable tool, such as tsung
. The problem with profiling two different code bases is that you're comparing apples to oranges. Things like memory consumption (declared to be greatly reduced in 13.x due to switching from strings to binaries) do count in real-world scenarios when you intend to host thousands of concurrent connections. Also note that programs like ejabbed
do really tend to be memory- and I/O-bound, not CPU-bound (the latter usually indicates you've managed to trigger a bug in the program or runtime) so mere profiling might not be indicative, only stress testing the staging or production server would.