4.3.6 GUEST CPU high load (after the guest tool installed?) : win7 64-bit host, Debian GNU/Linux 32-bit gutest

[ci-zephyurus]

Thank you for making the great software available. I have been using virtualbox for sometime. Today, I upgraded from 4.3.4 to 4.3.6.

After installing the guest tools, I noticed the guest system is so sluggish and I could not even start thunderbird mail client to show the initial screen. Or rather, I noticed the sluggishness when I noticed the thunderbird mail client to take forever to show the initial screen.

I noticed that the CPU usage goes up to 100% (I have configured two cores for VM, two of them are at 100%.) The load was 7.0+, very high.

When I checked the output of "ps -aef | grep thunderbird" it is in the list. But I could not even kill it using "kill -kill pid". "top" output does not point to anything special.

So I suspect something is very fishy.

Eventually I think the thunderbird process was killed because I noticed that strace -p pid no longer worked. But the 100% CPU usage lingered on.

So I suspect that the device driver(s) in the guest tool may be to blame.

I am attaching the log to this filing.

I set the priority to major since it is unusable in my setup. I just re-tried to reboot the machine after setting the # of core to 4 cores, but again it seems to take unusually long to boot, and so I canceled it, and reverting to 4.3.4 which worked just fine.

TIA

[ci-zephyurus]

Log of the run when I installed guest tool and noticed very slow operation and high guest CPU usage.

[ci-zephyurus]

I have narrowed down the problem to a background defragmentation program that runs under host Windows7 64-bit.

Auslogics Defrag 3.6.1.0 (free version)

I ran this in the background, and while this program touches .vdi file, suddenly the CPU jumped to 100%.

Obviously, when I reported the problem, the defrag was doing its job earnestly.

What puzzled me, though, is this: I would think whatever defrag program on the host OS does may lengthen the I/O latency time of the guest OS and NOT the CPU USAGE VALUE in the guest(?)

(Maybe, it is. It may be that my use of IDE driver may experience high CPU usage due to the long I/O latency time. Changing to SCSI may lessen this impact of defrag program on the host.)

Also, I *think* 4.2.xx series suffered less from such disturbance of defrag program if I am not mistaken. Maybe the new driver(s) in the guest tool or the new VBox itself have some kind of call back to handle the relocation of disk block on the fly, which is adding to the very high CPU usage? Anyway, now when I stop the background defrag on the host temporarily I could lower the CPU usage inside guest. (But when I reported the issue this morning, the CPU usage remained high for a prolonged time, and I had to shutdown the guest and restart VBox. Oh well, there could be an issue of re-loading driver as well?)

Anyway, after re-installing 4.3.4 and saw there is very little problem with it (after stopping the defrag on the host), I resumed defrag and saw the CPU usage within guest went up quickly. But then I could again stop defrag and CPU usage in the guest dropped.

So I re-installed 4.3.6 and tried to resume and stop defrag on the host. This time, by stopping defrag I could lower the CPU usage within guest OS which became high when I resumed defrag operation on the host. I have no idea why the high CPU usage persisted earlier.

Now that defrag on the host was causing the issue, my main questions/observations now are :

• Why high CPU usage in the guest instead of the report of long I/O latency time? (I am using xosview to look at CPU usage constantly.)

• 4.2.xx series seems to suffer much less from defrag interference.

TIA

[aeichner]

Please reopen if still relevant with a recent VirtualBox release.