[Xen-users] Which clocksource should dom0 be using?

[Andy Smith <andy@xxxxxxxxxxxxxx>]

Hi,

I've got a bunch of Debian lenny dom0s, and a couple of Debian
squeeze dom0s. The Debian lenny ones are using kernel
2.6.26-2-xen-amd64. The Debian squeeze ones are using kernel
2.6.32-5-xen-amd64.

Due to problems with seeing "clocksource/0: Time went backwards"
messages in Debian lenny, all my lenny dom0s and all my lenny domUs
have clocksource=jiffies.

Being a new deployment, I did not force clocksource=jiffies on my
squeeze dom0s, so they have the default clocksource=xen.

What I am now seeing with lenny domUs (clocksource=jiffies) on the
squeeze dom0s (clocksource=xen) is that their timer granularity
appears to be limited to 4ms. It is possible to observe this for
example when using ping or traceroute; all timings will be a
multiple of 4ms, with the smaller ones being rounded down to
0.000ms.

I did some searching and what little I can find on this matter says
that to fix it, the domUs should be using clocksource=xen. That does
fix the 4ms granularity issue, but then the "clocksource/0: Time
went backwards" issue is back.

My Debian squeeze domUs, which are using pvops kernels and don't
have their clocksource forced (so are defaulting to clocksource=xen)
don't exhibit either of these problems.

Now, most of my squeeze dom0s only have "xen" and "tsc" as available
clock sources. One of them only has "xen".

Is there a combination of settings that avoids these problems,
without upgrading the Debian lenny domUs to squeeze?

Is the recommended clocksource for dom0 "xen"?

Cheers,
Andy

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users