On Tue, Dec 19, 2000 at 03:03:43PM +0000, Thomas Lockhart wrote:
> o WAL fsync() log commits and cleanup (aggregate throughput is great,
> but every once in a while someone waits while the paperwork gets done.
> Waiting may be due to processor resource competition)
>
> o Underlying file system bookkeeping from the kernel. e.g. flushing
> buffers to disk etc etc.
I was going to suggest the same, but it's interesting that it happens
on reads as well. I can't tell for sure from the graph, but it looks
like it happens fairly consistently - every Nth time. I'd be curious
to see how this changes if you artificially slow down your loop, or
adjust your OS's filesystem parameters. It may give some more clues.
--
Christopher Masto Senior Network Monkey NetMonger Communications
chris@netmonger.net info@netmonger.net http://www.netmonger.net
Free yourself, free your machine, free the daemon -- http://www.freebsd.org/
