|
Hi, just some initial questions to
make things clearer for those who
can really answer you (I don't think I can, so others please
watch this
thread for information you can respond to)?
On 2/1/2013 11:07 AM, kovatsch janos wrote:
hello
we are using quemu for some automated tests for a few
years now to test various software.
initially it was decided that fedora was to be used, so
all of the test machines had fedora 12 x64 installed.
we are using quemu 0.11.0 (qemu-kvm-0.11.0) and cannot
upgrade it, so we are stuck with this version of qemu and fedora
12 because of the following reasons:
the test process goes as follows:
-a qemu image is loaded that has windows installed on it
(version doesn't matter, it can be xp,7-x86/x64), the operating
system is in a running state
-the test is ran for a few minutes, then the vm is
stopped
-it reverts to the image and starts a new virtual machine
How exactly (what commands to what program) are you using to do
the
revert (It seems you are using the libvirt wrapper, please
confirm).
What file format are you using for the virtual disks
(qcow2, qcow2
on top of other disk format, qed on top of other disk format, host
OS partition, LUN on external RAID box...).
How are you making the snapshots (qemu monitor vmsave,
libvirt
snapshot, EVM snapshot, RAID box LUN snapshot, other...).
How are you restoring the snapshots (qemu monitor, libvirt
snapshot,
copying in a fresh copy of the state files)
When you switch from Fedora 12 to Fedora 13, how do the
versions of
other related software (libvirt, Linux kernel, host disk file
system,
...) change?
Have you tried "mix-and-match" to check if it is one of the
other
programs that are slowing things down?
note that every virtual machine image has 1 gb ram
(some xp images have 768 mb)
this works very well on the given qemu version, but as
soon as we upgraded to later versions of fedora, after a few
days, there always appeared a massive slowdown at the step of
reverting and running the vm (it grew from 5 seconds to
sometimes over a minute)
the distribution does not matter as other distributions were
tested too
example:
fedora 12: revert to image and start running virtual
machine (5 seconds)->run test(1-2 minutes)->close vm(few
seconds)->restart the process
fedora 13 and later, first day: revert to image and start
running virtual machine (5 seconds)->run test(1-2
minutes)->close vm(few seconds)->restart the process
fedora 13 and later, second day: revert to image and
start running virtual machine (15-20 seconds)->run test(1-2
minutes)->close vm(few seconds)->restart the process
fedora 13 and later, after 5-7 days: revert to image and
start running virtual machine (60+ seconds)->run test(1-2
minutes)->close vm(few seconds)->restart the process
there is a nasty workaround for this slowdown: if we
create a new snapshot at the end of every day, we can keep a
daily constant 5-6 second revert time
methods to reproduce:
-install fedora 13 or later (x64)
-install qemu-kvm
-create virtual machine with windows xp/7
-create snapshot of a running machine
-start reverting and measure the time it takes to do the
revert (gettime->virsh snapshot-revert->gettime)
-after a few days there should be an increase in the time it
takes to execute the script
there is another way to avoid waiting days:
simply set the system time to 1 week before and start the
revert, it should take almost 2 minutes to do the revert
thank you
some links:
https://bugzilla.redhat.com/show_bug.cgi?id=618720
http://lists.gnu.org/archive/html/qemu-devel/2010-08/msg00133.html
https://github.com/torvalds/linux/commit/7972995b0c346de76
http://lists.gnu.org/archive/html/qemu-devel/2011-03/msg02645.html
Enjoy
Jakob
--
Jakob Bohm, CIO, Partner, WiseMo A/S. http://www.wisemo.com
Transformervej 29, 2730 Herlev, Denmark. Direct +45 31 13 16 10
This public discussion message is non-binding and may contain errors.
WiseMo - Remote Service Management for PCs, Phones and Embedded
|