Oracle 11g on Win2k8 - 100% guest CPU, 15% host

[J1mbo]

I have an odd issue with a new install of Oracle 11g on Windows 2008 running on ESX U1.

VM:

- Win2k8, 8GB RAM, 2 vCPU

- Disk 1 - 40GB, thick-provisioned, LSI-SAS controller

- Disk 2 - 250GB, thick-provisioned, paravirtual controller

Host is Dell R610, 2x X5570, hyper-threading enabled, 48GB RAM, no other guests running for this test.

Oracle was running perfectly - calibrate_io showing 6500 iops (EqualLogic storage).

Now running calibrate_io, the Oracle process is using 100% CPU resource as viewed from Windows task manager, but viewed from vCentre the VM CPU usage is perhaps 15 to 25% only. Eventually it reports 200 IOPS.

I reduced to 1 vCPU, but the same happens.

Running calc.exe to calculate 12345678! reports 100% CPU in Task Manager and 100% in vCentre as would be expected.

Last job to run normally was a large(ish) import. The VM has a snapshot (which because of the import is now 60GB). I have moved it between hosts and datastores with no differences.

Any help would be very greatfully received.

Thanks in advance!

[J1mbo]

For the record, the issue turned out to be just the snapshot, which was about 60GB existed due to a database import. Once consolidated the performance returned to normal.

The job impacted the performance of the calibrate_io task seemingly because the processing overhead firstly reduced the CPU allocation to the guest (hence the 100% guest level/15% host level thing) but secondly added great latency. Since the calibrate_io process was running with a 20ms latency limit specified, the queue wasn't being ramped at all hence the rubbish throughput.