Runaway CPU under redhat linux - RC2

E

emmby

Bit poster
Using Parallels RC2, OS X 10.4.6 and RHES4 U3, I've noticed that usually my Parallels process consumes a very high load when it's essentially doing nothing. CPU usages of 60%+ are not uncommon (as reported by the Activity Monitor on my dual core macbook pro). I don't believe I've seen the same problem using a Windows XP guest, although I use redhat much more frequently than Windows so I may just have not noticed it.

As an example, I just cold-started my VM about two minutes ago. The VM is running almost no services except for Oracle 10g and nfsd. top in the guest reports negligible cpu activity, but the host Activity Monitor reports 46% CPU going to the Parallels app.

Here's the interesting part. I can now use "poweroff" to halt my VM's kernel so that there is NOTHING running in the VM at all, but Parallels CPU usage remains at 53% after the "Power down" message from the kernel. If I then click the Stop button on the VM window the parallels app finally stops doing whatever its doing to my cpu.

Some things to note about my setup. Most of these are likely irrelevant but thought I would mention them just in case.

  • Guest: RHES4u3 running Oracle 10g, 256M memory
  • Host-only networking, static guest ip
  • no usb, sound, floppy, or pc speaker
  • High Acceleration, VT-x support enabled

Anyone seen similar issues? I seem to recall a bug against Windows guests regarding CPU usage when idle that was fixed prior to RC2. I wonder if perhaps this issue could be related.

Mike
 
I have posted here and filed a problem report for a similar problem with one application only under Windows XP Pro. This application sits idle, basically polling for key strokes and screen updates. It shows virtually no Windows activity (< 5%) but both CPUs on my Mac show usage between 65 and 75%. I close the app and the Mac CPU usage becomes almost nill. I have only found one app so far that has this behaviour.
 
That's interesting. It does sound different than the problem I'm reporting though, since the behavior I observe is independent of what's running within the guest VM. Or at least the behavior continues after the guest kernel has been halted anyway
 
I am thinking that there might be some kind of instruction that Parallels runs in specific cases that either has an incorrect time out parameter or something else that makes it run amoke. It is possible that, if this is the case, that your guest OS uses the specific instruction (or API) that my problem application is using.

The biggest simularity between our two cases is that it would see that the guest OSs think that they aren't busy at all but something is running amoke with the OS X processes. By the way, according to the activity monitor, it seems to be a kernel call in my case that is eating the cpu bandwidth.

Just a possibility. Would be interesting if anyone else has seen similar behavior.
 
You must log in or register to reply here.
Back
Top