[Qemu-discuss] Strange freezes in Win2008 Guest on recent hosted kvm

[Jakob Bohm]

Dear list,

We are having a strange problem with some virtual servers
hosted by a commercial provider who uses qemu+kvm.  We have
been working with them to try to find the root cause, but so
far no luck, and I am somewhat out of my depth with this deep
level of qemu/kvm internals.

So I wonder if someone here on the list could help me track
down the issue.

The symptoms are that at various times, one of the relatively
lightly loaded Guests suddenly experiences 3 problems almost
at the same time:

1. The disk queue length gets very large.

2. The Guest stops responding to pings.

3. The Guests clock stops.

After about a minute, the Guest springs back to life, but its
clock is behind by the duration until SNTP resyncs (It is set
to poll our local NTP server every minute).

This happens a few times a day, and each time, the primary
networked services on the Guest appear failed to the rest of
the world.

Hosts config (according to the provider):
   AMD Opteron 6128
   Enough RAM
   Two hardware NICs, one on live unfiltered IPv4 Internet,
      one on a LAN
   Local disks in RAID 1 configuration, shared with the other
      hosts as iSCSI.
   Linux kernel 3.1.4
   Qemu 1.0 with some unidentified patches (They won't tell).
   Providers management software.

Guests virtual hardware config (for each Guest):
   Two vcpus, each virtually 1GHz
   1.5GiByte RAM
   Two vNICs, eth0 connected to the Internet, eth1 to
      a VLAN between the Guests.  E1000 and virtio tried, most
      tests done with E1000.
   80GB virtual hard drive, backed by an iSCSI partition or
      LVM volume.  partition may or may not be on the same
      host machines.  IDE and virtio tried, most tests done
      with IDE.

Guests software config (for each Guest):
   NT 6.0 32 bit x86 (Windows 2008 Enterprise Edition, 32 bit)
   latest virtio drivers from
http://people.redhat.com/vrozenfe/build-26/virtio-win-prewhql-0.1.zip
   static public IPv4 and 6to4 IPv6 on eth0.
   static RFC1918 IPv4 and two static routable IPv6 addresses (in
      6to4 nets) on eth1.
   MS SQL Server, MS IIS etc.
   CYGWIN

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