It was just read-write vs sendfile vs map-write-munmap (vs the
"perfect" numbers of a just write). All the cases require the same
open-close code, so I didn't think that was worth testing because
it shouldn't change relative timings, just add constant overhead.
Strange... I am curious what the sendfile vs mmap (ie. don't do the
map/unmap parts around each iteration) case is. A lot of boneheads on
64-bit machines get killer specweb by mmaping readonly the entire
working set of the benchmark file set during the warmup run, and just
keep them mapped like this.
It's out of curiosity only, I prefer the sendfile() method personally
if the TCP bunching issue can be worked out to everyone's satisfaction.
Later,
David S. Miller
davem@dm.cobaltmicro.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu