Andrew Morton sent us a note stating that last
of the new API for putting processes to sleep missed
an important objective of that work. The new interface is nice, but what
he was really setting out to do was to improve wakeup performance. The new
code removes waiting processes from the wait queue immediately at wakeup
time, rather than letting the processes remove themselves whenever they get
around to it. The result is that subsequent wakeups, if they come quickly,
will run faster because they do not need to deal with processes that have
already been awakened.
We also mentioned, last week, a posting on the leading-edge features used
in the TPC benchmark results posted by HP. Lest anybody think that HP was
using a highly patched, special-purpose kernel, they have posted a followup stating that a stock Red Hat
kernel (from Advanced Server 2.1) was used in the benchmark runs.
Ingo Molnar's new process ID allocator - and the objections to it - were
covered last week. Ingo posted a new
version of the patch which addressed some of the complaints, and which
was to Linus's liking; it was merged into the 2.5.37 kernel.
to post comments)