<html dir="ltr">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style id="owaParaStyle" type="text/css">P {margin-top:0;margin-bottom:0;}</style>
</head>
<body ocsi="0" fpstyle="1">
<div style="direction: ltr;font-family: Tahoma;color: #000000;font-size: 10pt;">Hello,<br>
<br>
we have an up to date ovirt 3.4 installation. Inside we are running SLES11 SP3<br>
VMs (Kernel 3.0.76-0.11). After live migration of these VMs they all of a sudden<br>
do not react any longer and CPU usage of the VM goes to 100%.<br>
<br>
We identified kvm-clock source to be the culprit and therefore switched to another<br>
clocksource. We ended with hpet but are not happy with that as our inital goal<br>
was to use the more simple designed TSC clocksoure. <br>
<br>
The reason behind that is the question I have for you experts.<br>
<br>
Our hosts all have the constant_tsc CPU flag available. Just to mention these<br>
are not identical hosts. We have a mix of Xeon 5500 and 5600 machines. E.G.<br>
[root@colovn01 ~]# cat /proc/cpuinfo | grep constant_tsc | wc -l<br>
8<br>
<br>
When we start the VM the client sees TSC as available clocksource:<br>
<br>
colvm53:~ # cat /sys/devices/system/clocksource/clocksource0/available_clocksource<br>
kvm-clock tsc hpet acpi_pm<br>
<br>
After the first live migration to another host that also has constant_tsc (see above)<br>
that flag is lost inside the VM.<br>
<br>
colvm53:~ # cat /sys/devices/system/clocksource/clocksource0/available_clocksource<br>
kvm-clock hpet acpi_pm<br>
<br>
Any ideas?<br>
<br>
Markus<br>
<br>
<br>
</div>
</body>
</html>