Guest slowing down host

[WildDaemon]

Hi guys/gals,

As the title describes I'm experiencing that the guest is slowing down the host system. What you usually see in the bug-tracker is that the host is slowing down the guest, for me that is NOT the case. I've noticed this thanks to my G19 keyboard which displays the clock. The second hand on the clock was progressing too slow and that caught my attention. This problem has been going on for months now (hoping it would be solved) and today I narrowed it down to a specific VirtualBox build using benchmarks/tests/<name it whatever you want>. I am going to describe what I did, what helped and what didn't help.

FYI: The tests I've run are all related to AMD-V as my machine is a Phenom II X4 940 with 4GB of RAM running Windows 7 (x64) on a SATA II HDD with 16MB of cache. Please do not tell me the problem can be solved by switching Cool 'n' Quiet on or off, this doesn't solve the problem at all as it's not hardware or BIOS related in any way. It's not related to a specific Windows 7 installation either because I've also tested it with a clean install of Windows 7 (32 bit) without any software installed whatsoever. I'm using the official binaries which can be found on the Downloads page. I'm NOT using the Open Source Edition nor am I using the Extension Pack. No offense but please don't be a smart* and tell me it related to anything I've just said it isn't. I've mentioned it and will mention it again: the problem I have is from a specific build.

The way I tested was using the Windows "Date and Time" window together with a mobile phone which shows the clock on it's display. With the non-laggy build (3.2.10) both clocks run perfectly synchronous. The newer versions (3.2.12 and upwards) show a lag unless the guest creates heavy disk/CPU usage for the host (e.g.: booting the OS or installing Guest Additions). I've tested with AMD-V and I/O APIC on and off for single-core and quad-core use (were applicable) on a 32-bit guest (Windows, Fedora 14 and Ubuntu 9.04) without Guest Additions. A 64-bit guest does not seem to improve the situation. These are the results I've gathered:

Boot times (x86 and x64 together):

• VBox 3.2.10 (no lag)
  • Single core
    • AMD-V disabled: 45 - 47 seconds
    • AMD-V enabled: 15 seconds
  • Quad core (using I/O APIC)
    • AMD-V enabled: 13 - 15 seconds
• VBox 3.2.12
  • Single core
    • AMD-V disabled: 45 seconds
    • AMD-V enabled: 17 - 18 seconds
  • Quad core (using I/O APIC)
    • AMD-V enabled: 13 - 16 seconds

As you can see it doesn't matter which build you install and it still looks good, but then lets see what it does to the host clock..

Seconds processed:

• VBox 3.2.10 (no lag): advanced 60 out of 60 seconds for both the x86 and x64 OSes
• VBox 3.2.12
  • Single core
    • AMD-V disabled
      • First minute: advances 52-53 out of 60 seconds (1 second takes about 1.142 seconds!)
      • Second minute: advances 34-37 out of 60 seconds (1 second takes about 1.692 seconds!)
    • AMD-V enabled
      • First minute: advances 40-41 out of 60 seconds (1 second takes about 1.481 seconds!)
      • Second minute: advances 34-36 out of 60 seconds (1 second takes about 1.715 seconds!)
  • Quad core (using I/O APIC)
    • AMD-V enabled
      • First minute: advances 39-40 out of 60 seconds (1 second takes about 1.519 seconds!)
      • Second minute: advances 35-36 out of 60 seconds (1 second takes about 1.690 seconds!)

Now these results are stunning, right? What it translate to is that every 5 minutes you lose 2, which pretty much means that every hour the clock is 24 minutes behind! For what it's worth I think it relates to the "VMM: use new VT-x feature to keep the guest from hogging the CPU" feature that has been added in v3.2.12 build 68302 which might not work for AMD.

HTH,

[wwirthtorres]

I have been experiencing this same issue on v4.2.6 r82870 My Host: AMD Phenom 9950 Quad-Core, Windows 7 Pro 2 Guest OS: Win Server 2008 R2

When one or both guests are running, I lose approximately 12 minutes per hour on the Host clock. The more odd part of the behaviour is that once all process threads have completed running after closing down all guests and Vbox, the clock resets to the correct time.

[Frank Mehnert]

So what you say is that the clock of your host operating system runs slower if you run a VM on that host, did I understand this correct?

I would like to see a VBox.log file of such a VM session, mainly to see the configuration of your VM.