Hi,

I've been running a UML-based Linode where the host kept my Linode's time nicely in sync.

I've recently (well, about 22 days ago) changed to a Xen host and to kernel 2.6.27.4-linode14 on my Xen Linode. Since then, I've noticed that my time is drifting away from the correct time.

I've tried having a look at /proc/sys/xen, but there doesn't seem to be a xen directory in the /proc filesystem where one can set wallclock settings.

Can I still expect my guest time to be "kept" by the host, or should I be keeping my own time using NTP? Or is the host maybe drifting? (I'm on dallas101).

--deckert

hm... I also converted from UML recently, I too seem to have picked up a noticeable offset. Nice catch.

Straight from the horse's mouth:

(Context: emag was asking if he should run NTP on his Linode)
http://thegrebs.com/irc/linode/2009/01/21#15:40

We suggest running "Latest 2.6" for now, unless you want weird things to happen...

-Chris

Indeed... I was about 2 seconds off yesterday, and looking at my logs it looks like I converged, but then started to drift faster than NTP could keep up.

Hmm. I've been running 2.6.27-4-linode14 since the day Chris announced it. Nothing weird has happened, and ntpd keeps my clock in sync just fine.

Who knows? Maybe I'm just lucky, but I'm sticking with this kernel. It's noticeably faster.

Looking at my logs some big adjustments have been required (> 100 ms), but I don't think I've ever expected terribly precise timekeeping from a virtualization solution. What I'd be worried about (this can happen with VMWare without VMWare tools support) is jumping to such a big offset that NTP gives up, and to a certain extent it's annoying if you disagree significantly for dealing with e-mail and stuff. That hasn't happened, but if this is anything like VMWare it is somewhat dependent on load, so we'll see I guess. Under a second isn't a huge problem for me.

I had some IPv6 issues with 2.6.18 and longer term I'll be wanting to use BTRFS, so I too will be sticking with the newer kernels. I think ultimately newer kernels are either adequate or they're not, because 2.6.18 forever is not a long term option. If this is the worst we see with newer kernels, I'll be pretty happy.