Wireshark-dev: [Wireshark-dev] speed optimizations
From: "- Aucun -" <[email protected]>
Date: Thu, 31 Oct 2013 12:03:27 +0100

Years ago for our internal use I made a lot of speed related changes to
wireshark (then ethereal).

Some of these changes are currently in wireshark but many aren't.

Lately I'm looking at qt work and I'd like to move our internal tool to it.
A simple merge is DOA, vs our wireshark version the diff is a 100MB patch :(,
thus it's simpler to more or less rewrite them, they aren't that big, and it's
the right time to try to move them in wireshark code.

Of course a proper patch ready for commit in wireshark tree is more work than
a quick and dirty one good enough for only our usage. 

Are wireshark's developers interested in reviewing and committing these patches?

There's roughly four set of patches.
-1 Base on Yami's work "snapshot feature: apply filter upon previous displayed
packets only" (without the SAT, only using a true table). help a lot when
users are building filters incrementally or move a lot between them.
-2 When filtering don't dissect packet if not necessary. Help a lot, can speed
filtering up to 2 orders of magnitude (twice to three time as fast common).

-3 a lot of small ones which only make sense once 2 is in, for example when
you're filtering at 300 000 packets/sec, one lseek/read per packet hurt.

-4 rewrite packet dissectors, AOT and JIT(*), we are using an internal tool I
can't share (too buggy) but it could be done with either LLVM or ugly
preprocessing at compile time for AOT.

I'm willing to work on pushing 1,2,3 in wireshark.

It's more complicate for 4 as it's working for us (ie I don't have a budget
for a big rewrite using LLVM). 

Any takers?