different CPU load in idling (W2k << Linux2.6 <W98)

mimi

Member
I noticed that there are quite large differences in the host CPU's load for different guest OSs when they are in idle state. The values monitored by top command were something like:
Windows2000 3~5 %,
Linux2.6 ~25% (Scientific linux4.2, a free version of RH Enterprise),
Windows98se ~30%.

I don't care for Win98, however, such a large load for linux disappointed me.:confused:
I remember that a typical load for Win2K was also about 30% in earlier betas. So can I expect lower CPU load for linux in later betas:) ?

MBP 2G, 2Gmem, PW2.1 beta6.
 
mimi said:
I noticed that there are quite large differences in the host CPU's load for different guest OSs when they are in idle state. The values monitored by top command were something like:
Windows2000 3~5 %,
Linux2.6 ~25% (Scientific linux4.2, a free version of RH Enterprise),
Windows98se ~30%.

I don't care for Win98, however, such a large load for linux disappointed me.:confused:
I remember that a typical load for Win2K was also about 30% in earlier betas. So can I expect lower CPU load for linux in later betas:) ?

MBP 2G, 2Gmem, PW2.1 beta6.

Did you select the correct OS in the PW Edit dialog? This can make a difference.

However, to an extent this is to be expected: an "idle" operating system is still doing _something_, and how each OS handles idle varies. Most VM systems (I assume, including PW) try to detect these idle states and switch away from the OS when this occurs. For example, deep in the bowels of WinXP is (in effect) a system idle thread that just executes a HLT instruction to stop the CPU when nothing else is happening. PW can detect this (in fact, OS X can also) and switch away from the thread. However, some system such as earlier Windows systems don't do this can can always appear "busy". This is why you see such different % times during so-called "idle".

--Tim
 
Errata: CPU usage in idle state (W2K << Linux <<< W98 )

Tim,
thank you for your reply and I am sorry I wrote wrong value for Win98.
They should be:

Windows2000 3~5%,
Linux2.6 ~25%,
Windows98se ~100% (it means one of the cores is fully occupied by parallels!).

I know there are many background processes even in idle state, the cpu usage
in linux guest indicates 99% idle. So I concluded that parallels does some redundant works
for linux guest. I wish Parallels team provides some tweaks for this problem.

Here are the results of top command on the host for different guest OSs. I measured
just after the first login from a cold boot of guest. I have installed parallels tool for Win2k
and win98.

Windows2000:
PID COMMAND %CPU TIME #TH #PRTS #MREGS RPRVT RSHRD RSIZE VSIZE
2110 Parallels 0.0% 0:00.00 1 8 101 1.96M 12.1M 644K 92.6M
2109 Parallels 4.1% 16:14.10 13 214 475 57.3M 639M 185M 1.24G

Linux2.6:
PID COMMAND %CPU TIME #TH #PRTS #MREGS RPRVT RSHRD RSIZE VSIZE
2110 Parallels 0.0% 0:00.00 1 8 101 1.96M 11.0M 644K 92.6M
2109 Parallels 22.7% 2:56.13 17 182 450 58.1M- 371M 251M- 999M-

Windows98se:
PID COMMAND %CPU TIME #TH #PRTS #MREGS RPRVT RSHRD RSIZE VSIZE
2110 Parallels 0.0% 0:00.00 1 8 101 1.98M 11.7M 644K 92.6M
2109 Parallels 102.0% 14:02.39 17 207 471 59.2M 213M 103M 1015M

host:
MBP 2G, 2Gmem, PW2.1 beta6.
 
You should flag this as a bug in PW. However, you might also want to look around and see what it is you have running in Win98SE. Anti-virus? Screen-saver? Even innocent tools might have done something bad. Win98 software was generally pretty awful when it came to hogging the cpu.

--Tim
 
mimi said:
I noticed that there are quite large differences in the host CPU's load for different guest OSs when they are in idle state. The values monitored by top command were something like:
Windows2000 3~5 %,
Linux2.6 ~25% (Scientific linux4.2, a free version of RH Enterprise),

Hello Mimi

I think this is also a function of the guest operating system running.
I have Suse 10.0 as a Linux guest on a Mac Mini core duo,
2 GB ram total, 1 GB ram dedicated to the parallels VM.

Screenshot: http://www.mission-base.com/peter/screenshot.jpg

OS X shows a load of about 10%, ( upper and lower right of the screen shot )
while the Linux guest shows 0% usage. ( upper left of the screen shot )
10 % loss is not bad, I think, considering that the linux machine is
fully visible and functioning as a server on my network.

Greetings!

Peter
 
.
The Windows idle bug was reported and fixed some time back. You will have to do your own searching in forums for that.

Doing a bug report now for Linux would be the way to go.
.
 
expect less cpu usage for linux guests

thank you for some replies.
I used bare Win98SE without any further installations. I suppose the kernel of Win98 is quite different from that of Win2K/XP. So I would not care about Win98 anymore.
I know the idle problem in Win2K/XP was discussed sometime ago and then PW team tweaked to solve the problem in beta2~4. Therefore I would expect such tweaks for Linux, too. I think 10% is still not small enough for me.
 
As reported in another thread, I am experiencing moderately high core usage when idling in XP Home SP2 guest using Beta 6.
 
same with RC, unfortunately

The problem still exists in RC, unfortunately.

Win2K:
382 Parallels 0.0% 0:00.00 1 8 89 1.92M 18.4M 984K 97.9M
381 Parallels 4.3% 2:10.47 10 127 398 20.7M 697M 203M 1.12G

Linux2.6:
382 Parallels 0.0% 0:00.00 1 8 89 1.94M 19.8M 984K 97.9M
381 Parallels 28.8% 3:33.20 19 215 481 22.1M 439M 232M 929M

Win98SE:
382 Parallels 0.0% 0:00.00 1 8 89 1.94M 19.9M 984K 97.9M
381 Parallels 102.6% 7:31.88 17 220 473 22.7M 252M 82.3M 929M

PW2.1 RC B1832, MBP 2G 2Gmem.

About 30% of CPU usage for Linux guest is preserved even after the completion
of a shutdown procedure ( till manually "power off" the VM). It seems that the detection
of HALT status is not perfect.
 
RC2: improved for W98

Since RC2 is available now, I measured the cpu load again for three OSs.
As stated in the announcement, the 98% load issue in Win98 has been solved.
I am still wondering why win2K VM is so light compared with others. Linux VM
consumes 28% load even after shutdown.


Win2K:
327 Parallels 5.1% 1:42.27 12 145 454 23.6M 743M 157M 1.22G

Win98:
327 Parallels 27.6% 3:53.74 17 213 495 24.5M 317M 83.8M 1003M

Linux2.6:
327 Parallels 27.9% 6:09.07 19 228 514 24.8M 481M- 235M- 1005M-

Linux2.6: Power-down (VM is running)
327 Parallels 27.8% 6:43.75 19 228 510 24.8M 481M 236M 1005M

No VM is running
327 Parallels 0.5% 7:06.58 4 196 447 23.5M 188M 54.2M 695M


MBP 2G, 2G-mem, PW 2.1 RC2, 1.2G mem at max. is allocated for PW.
 
GA: still >25% CPU load in idle

Since GA is available now, I checked CPU loads of three different VM. Unfortunately,
no improvements from RC2.

Linux2.6:
293 Parallels 25.8% 1:48:31 19 209 660 22.8M 320M 215M 995M

Linux2.6: (Power off)
293 Parallels 25.8% 1:49:23 19 209 656 22.3M 329M 217M 995M

Win98:
293 Parallels 24.1% 1:52:00 17 208 651 22.5M 156M 65.6M 993M

Win2K:
293 Parallels 3.7% 1:56:37 13 212 653 14.8M 585M 128M 1.22G
 
constant said:
.
The Windows idle bug was reported and fixed some time back. You will have to do your own searching in forums for that.
That's not been my experience. The GA release seems to have addressed part of the problem. Previously, SMB sharing causing PWS to spin out and pull 10-25% just sitting there. In the GA, it's about 5-10%. If I turn off SMB sharing, it drops to < 5%. This is much better than the RC's but there is still something I can't nail down. PWS also does not release all memory after exit but that's a topic for another thread...

Don't get me wrong, this is by far better than VPC ever was and I am thoroughly happy with PWS.
 
I'm experiencing the same symptom

I'm experiencing the same ~25% busy time with the same idle guest OS, Scientific Linux 4.2. Just another datapoint. My system is a Macbook Pro 1.83Ghz, 1GB memory. The guest has 384MB assigned to it.
 
Back
Top