On Sun, Oct 24, 2010 at 10:56:40PM +0000, David Holland wrote:
> Anyway, ISTM that writing from the mmap buffer in say 64K chunks would
> retain nearly all the advantages and get rid of the latency problem.
The way the code is currently written it only uses mmap(2) for
files smaller than 8MB anyway. Your suggested change would require
more work than reducing the size of the mapped memory.
There is also the problem that the overhead per call to mmap(2) or
munmap(2) is high on some platforms, IIRC alpha is one of them.
Changing the code as you suggested above might therefore impact
performance on some ports.
Kind regards
--
Matthias Scheler http://zhadum.org.uk/
