superlifter design notes and a new proposal

[Martin Pool]

On  6 Aug 2002, Martin Pool <mbp at sourcefrog.net> wrote:

> Fast networks are a significant use case that ought to be considered.
> On GigE, having low processing overhead and being able to use things
> like sendfile() to transmit data are going to be enormously more
> important than the length of the header.

... and that kind of high-speed processing will be substantially
faster if the data is word-aligned, which implies an word-sized header
and padded data.  

I did some measurements a couple of years ago for aligned vs unaligned
memcpy, and it was very significant: >10% in CPU time at least.  This
will carry over to sendfile() as well.  It applies on Pentiums and
will be even greater on any other architecture.

I'm not particularly saying that we need to optimize for this
particular case, but rather that it is an example of why XDR is
designed the way it is.

-- 
Martin