[lwip-devel] Purpose of pool memory allocator

[lukem . lwip]

I am beginning to question the need for a pool allocator in lwIP, so I
thought I'd see what others thought about it.

Has anyone done any profiling which shows that lwIP spends a significant
amount of time allocating buffers? My own profiling shows a fairly tiny
amount (<0.1%) of time is spent allocating packets anyway, which is hardly
worth optimising for.

The other case, which could be a concern for embedded systems, is the
space overhead of a general purpose allocator. Does anyone have a feel for
how much overhead there would be if we switched to a general purpose
allocator for everything?

Now, there are a few advantages to getting rid of the pools
* Simplify the codebase
* Reduce the size of the compiled binary
* Potentially reduce the size of BSS
  - removing fixed allocation partitions between pool and ram types
    could allow the system to work under tighter memory constraints
* Saves me adding another default to the new buffer management code :)

--
Luke