Discrepancies between Oprofile and vmstat

[Styner, Douglas W]

We are using an OLTP-type workload on a fully saturated system and have begun seeing a discrepancy between the output of oprofile and vmstat on upstream kernels that was first seen with the 2.6.27 kernel.

Workload:
OLTP database workload using QLogic FC cards with MSI-X enabled doing ~50,000 IOPS

Oprofile (v0.9.3) collection methodology:
opcontrol --init
opcontrol --vmlinux=/usr/src/linux-`uname -r`/vmlinux
opcontrol -e=CPU_CLK_UNHALTED:80000 -e=LLC_MISSES:6000
opcontrol --start-daemon
sleep 5
opcontrol --start
sleep 60
opcontrol --stop
opcontrol --save=$1
opcontrol --deinit

Function classification:
Modules vmlinux and *.ko = kernel, all others = user.

Results (2.6.30-rc3 kernel):
vmstat reports ~75% user, 25% kernel
oprofile reports 70.3808% user, 29.6142% kernel 

Using another profiling tool to confirm, we see 74.784% user, 25.174% kernel.

Doug Styner
douglas.w.styner@xxxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/