Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=iso-8859-1 On a VM, it's needed to use an alternative for native cpu timing (-e = cpu-clock), but you've done that, otherwise you would see nothing. What is your kernel version? A quick peek on the internet shows it seems = to be a clock-source problem with older kernel versions. I haven't seen = this with 2.6.32+ Frits Hoogland http://fritshoogland.wordpress.com frits.hoogland@xxxxxxxxx +31 6 53569942 On Jul 12, 2013, at 8:08 PM, Niall Litchfield = <niall.litchfield@xxxxxxxxx> wrote: >=20 > On Jul 12, 2013 6:28 PM, "Frits Hoogland" <frits.hoogland@xxxxxxxxx> = wrote: > > > > It depends on where the time is spend. Use perf top or record to > > measure functions for threaded and non threaded performance. >=20 > I'm so glad you said that :). Consistently when I do this (on a vm) = the greatest single sample count is for acpi_pm_read. There's no good = explanation I've found for this. I have some ideas but 30% of samples in = this call compared with <5% in the various qer* Oracle function calls = seems wrong in a swingbench test. I suspect I'm missing the idiots guide = to perf/linux kernel functions.. >=20 -- //www.freelists.org/webpage/oracle-l