Large file replication uses multiple HTTP PUT requests for ranges of the
content, rather than just doing a PUT for the entire file contents at
When we tested this with nginx, it got it completely wrong, and
subsequent writes overwrote the first chunk..
Maybe this is the issue that you're seeing? I'm unsure if this has been
fixed in latter versions of nginx or not yet.
lighttpd doesn't suffer from this problem, but it limited to less than
2Gb (signed 32 bit value - don't ask me why it's signed!) on 32 bit
machines unless you modify the source code.