https://bugzilla.novell.com/show_bug.cgi?id=406298
User nice@titanic.nyme.hu added comment
https://bugzilla.novell.com/show_bug.cgi?id=406298#c4
--- Comment #4 from Tamás Németh 2008-07-21 14:32:57 MDT ---
Hi guys!
Sorry for being so disappointed. So:
The infrastructure is composed of Sun Blade X6250 servers, each with two
qla2xxx fibrechannel cards, which are connected through two FC fabrics to a Sun
StorageTek 6140. So, the storage is fibrechannel, with MD based multipath
access. See my bugreport about multipath-tools (which is still broken in 110
but in a different way):
https://bugzilla.novell.com/show_bug.cgi?id=397119
More about the StorageTek 6140:
https://bugzilla.novell.com/show_bug.cgi?id=398536
I have to use openSUSE 10.3, because the X.org videocard driver for these
servers is severely broken in 11.0. (And the 64 bit -xen kernel of openSUSE
11.0 is even unable to boot on these servers (it waits for the root partition
to appear in vain), while he 32 bit version works.) So, I tried 32 bit 10.3, 64
bit 10.3, 32 bit 11.0 and 64 bit 11.0, but decided to use 32 bit 10.3, because
at one point i ws interested in using the newest vanilla kernels as domU, and
they don't have Xen support for 64 bit. (The vanilla kernels' Xen domU support
is also broken, for example using the ballon driver causes immediate crash for
me, beside being unstable under heavy load.) It seems to me that the vanilla
kernels can run only on Xen 3.2.1, so i replaced the xen packages of openSUSE
10.3 with the newest xen, downloaded from xen.org and and compiled by me. So:
1: domU and dom0 is 32 bit openSUSE 10.3. Dom0 runs on the xen.org kernel, domU
runs on the suse kernel, compiled by me, to be 100Hz, non-preemptive, etc.
However, thee hypervisor is 64 bit, since the servers have 32GB of memory.
2: Fibrechannel (see above)
3: I can ping domU, and even Xvnc comtinues to operate (the "screen" of domU
changes), but i cannot start new programs. Sadly it includes the fact that i
can't even run dmesg. When i destroy and restart the domU, i cannot find
anything about the crash in /var/log/messages.
4. I had had the impression that time syncing issues are related to the kernel
preemtion model and timing frequency:
https://bugzilla.novell.com/show_bug.cgi?id=344877#c7)
This is my reason to recompile the kernel. (BTW why don't you ship the xen
kernels with similar settings?)
My other impression was, that the openSUSE xen kernel (at least the 32 bit
versions) are broken when running on multiple vcpus:
https://bugzilla.novell.com/show_bug.cgi?id=350051
https://bugzilla.novell.com/show_bug.cgi?id=343181
But this migrating issue now seems to be independent from vcpu number,
preemption model and timing frequency.
Answering your question: the dom0 systems are synchronized by ntp daemons.
Today the hypervisor, every xen tool and the kernels in both dom0 and domU are
those from xen.org, and EVERYTHING works fine, but without kernel updates,
special SuSE patches, and apparmor.
Maybe I will have the opportunity to change to SLES and XenServer.
--
Configure bugmail: https://bugzilla.novell.com/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.