Slow boot / high CPU load / snapshot delete stalls on 12 core Family 10h Opteron Ubuntu 10.04

[Jens Wilke]

I have some "multiple" strange behaviours on a new host system.

Parameters: Host: Ubuntu 10.04 aka Lucid 64 Bit, Kernel 2.6.32-31 Guest: Ubuntu 10.04 aka Lucid 32 Bit, Kernel 2.6.32-31 Virtualbox Version 4.0.8 AMD Opteron(tm) Processor 6174

I also did test with 11.04 and Kernel 2.6.38-8, and experimented with various guest options (ioapic, acpi, etc.). Overall, the behaviour keeps identicall.

Observations:

1. Slow output on the UART during boot.

2. The boot is slow. The process takes 55 seconds when 12 cores are enabled and 10 seconds when only 1 core is enabled.

3. When switching to multiple CPUs in the guest the UART output is corrupted during the initial boot phase.

4. When deleting a snapshot it stalls at 90% and the following processes run at the given CPU %, 63% VBoxSVC, 62% VBoxHeadless, 29% VBoxXPCOMIPCD. I wait for 20 minutes, no change. Then I aborted via rebooting the system.

The attached logs concentrate on the boot of in 10.04 with identical VM parameters, I did the following comparisons:

12core: All cores are enabled

1core: disabled all but one cpu with: for I in 1 2 3 4 5 6 7 8 9 10 11; do echo 0 > /sys/devices/system/cpu/cpu$I/online; done

downcore: Enabled BIOS Downcore setting to 1 core, which gives me 2 cores in reality.

1core and downcore have fine behaviour. So when the host system is booted with only 2 cores enabled, everything is fine!

[Jens Wilke]

Additional finding, which may be related or not. Kernel oops in host during guest operation while settings cpus online/offline.

[Jens Wilke]

Additional finding, which may be related or not. Kernel stacktrace during guest boot in a 11.04 / Natty host.

[Jens Wilke]

Update: Ticket can be closed, we switched to KVM for virtualization in the data center.