Load average spikes during merging?
Scott Lystig Fritchie
fritchie at snookles.com
Tue Aug 23 15:45:19 EDT 2011
Jeff Pollard <jeff.pollard at gmail.com> wrote:
jp> Overnight we do some data collection that is stored in Riak, and
jp> just last night we had one of our server's load spike very high and
jp> just drop back down to more acceptable levels. You can see a graph
jp> of it here.
Jeff, all that load average tells you is a weighted average of how many
processes are runnable: running or ready to run but no CPU is
available. Linux also seems to include processes in disk wait state,
see http://en.wikipedia.org/wiki/Load_(computing). I don't have any
specific knowledge of a Linux kernel's treatment of pthreads in
computing the load average, but it wouldn't surprise me each thread in
disk wait were included.
Riak will start a pool of Pthreads for I/O work, see the +A flag in
/etc/riak/vm.args (default Linux path), e.g.
## Enable kernel poll and a few async threads
If you're interested in better measurements, "iostat -x" will give you
much more useful information about CPU & disk usage than load average
will. Intervals of 5 seconds are about the largest worth watching;
anything larger loses too much in the average. 1 second is ideal,
More information about the riak-users