In message <199610232244.SAA27339@caip.rutgers.edu> "David S. Miller" writes:
: Therefore my micro-optimization to memcpy/csum_partial_copy which gave
: me 4 or 5 MB/s faster bandwidth over pipes and sockets is a SMALL
: improvement?

That's not a micro optimization.  That's optimizing a critical part of
the code path shown to be a bottleneck.  The system call overhead, on
the other hand, is likely in the micro range because it is generally
insignificant on most machines.  Datacoping happens all day long, and
is a source of major slowdowns and should be optimized.

It is hard to communicate with people using different definitions of
"micro".  Most of this thread is an emotional reaction to highly
charged words that revolve around the different definitions of the
things people are talking about.

David, your work has value and is useful.  However, just because you
and other folks with Linux have done things that are a big performance
increase (the data copy path, for example), does not necessarily mean
that everything you do will necessarily have that same result (I have
my doubts about how much faster systems with a minor improvement to
the system call overhead really helps, for example).  While you may
not be intending to say that, that is how you are coming off in this
tread.

Warner