performance: the library is too slow for production use

[Didier Barvaux]

The ROHC library was mainly designed with features in mind, not speed. While not very slow, the library is not ready for production environments. There is room for making the library more efficient.

This bug is for tracking changes that enhance the library speed.

[Didier Barvaux (didier-barvaux)]

Change target release from 1.3.0 to 1.4.0.

[Didier Barvaux (didier-barvaux)]

Check if using ASM for computing IP checksum is still a benefit. Do so for several targets: x86, x86_64, ARM, PPC...

[Didier Barvaux (didier-barvaux)]

Check if partial computing of IP checksum (remove old values, add new ones) is faster than computing the whole new checksum.

[Didier Barvaux (didier-barvaux)]

Check if -O3 improves speed without breaking the library.

[Didier Barvaux (didier-barvaux)]

Avoid memory allocation in critical paths.
Avoid payload copies if possible.

[Didier Barvaux (didier-barvaux)]

Use a hash for efficient search of contexts inside compressor: SipHash [https://en.wikipedia.org/wiki/SipHash] might be a great candidate.

[Didier Barvaux (didier-barvaux)]

Bench Clang vs. GCC

[Didier Barvaux (didier-barvaux)]

Bench the -fstack-protector-strong compiler option introduced by GCC 4.9.

[Didier Barvaux (didier-barvaux)]

Check if a LTO build improves speed without breaking the library.

[Didier Barvaux (didier-barvaux)]

Bench static vs. dynamic link.

[Didier Barvaux (didier-barvaux)]

try with native versions of rohc_ntoh[32|16] and rohc_hton[32|16]