Hi all.
I have a Windows 7 SP1 guest (with latest updates) installed in a virtual machine that is hosted on Ubuntu Maverick. Both host and guest are 32-bit operating systems. When the guest is running the system performs a lot of IO (the HDD LED is very bright) even for small operations and responds slowly. Doing something heavier, like starting Visual Studio or MS Expression Blend inside the guest takes around 5 minutes. Opening a XAML file or building my solution also takes unproportional amount of time.
I cloned the virtual HD and ran it under VirtualBox and it displays no such problems, so I believe the problem lies somewhere in the VMWare VM configuration. I wish to enjoy VMWare's ability to accelerate 3D graphics, so I have to stick with VMWare. Has anybody encountered such problems? Any solutions?
Some host info:
Core i5 520M
4GB of RAM (I have a PAE kernel, so it can access all of it)
Ubuntu Maverick (10.10)
500 GB, 4500 RPM hard drive. Has 277GB Free.
Guest info:
2 Virtual CPUs
2GB of RAM
100GB Virtual hard disk (not pre-allocated)
Another 10GB, pre-allocated virtual hard-disk for the windows swap file
I also added some advanced options that I saw in another site, that could potentially solve the problem but they didn't help (and didn't make anything worse too). You can find them at the end of the attached VMX file.
oops - you are right - I misread
numvcpus = 2
cpuid.coresPerSocket = 2
if you had 4 cores it would be
numvcpus = 4
cpuid.coresPerSocket = 2
about shrinking ...
a healthy vmdk should not have any of this entries in a vmware.log
Command READ(10) took 3.173 seconds (ok)
Command READ(10) took 3.246 seconds (ok)
Command READ(10) took 3.206 seconds (ok)
Command READ(10) took 3.618 seconds (ok)
eaither shrinking is long ago - or something else is wrong with your disk-system
where did you find those options ? - on my site ? (sanbarrow.com)
please attach a vmware.log
BTW
prefvmx.useRecommendedLockedMemSize = "TRUE"
should go to the config-file - not the vmx
also which filesystem do you use for /tmp ?
About the options - yes, from your site
About the option - I put the setting in the config file (you meant /etc/vmware/config, right?) and it didn't help nor made anything worse.
I attached the vmware.log file of a short session where windows loads, then I start visual studio, open my solution, close VS, and then open blend and close it.
In addition, I checked the Windows 7 resource monitor in the guest and noticed that both visual studio and blend use most of the IO when working. A small amount of IO went on the windows page file, but it was insignificant. However the IO bandwith was not larger than 3.5MB/sec. On the host I can copy files at around 30-40MB/sec, so it seems like the guests IO bandwidth is at most 10% of the host's capability.
About my /tmp, it's on the same partition with the rest of the system (/) and I use ext4.
there are a few things I do NOT like in your log
1. you use a VM with 4 cores - thats the same amount of cores as your host has - not ideal - try with 2 cores
2. your scsi0:0 disk has lots of delays - I doubt it will survive long if you keep it like this
first aid: shrink the scsi0:0 vmdk - or better - convert it to preallocated format
some vmplayer versions had problems with excessive fragmentation on ext4 - I am no Linux guy - please check if your systems suffers from heavy ext4 fragmentation
please do not ignore my warning - if you continue to use your VM as it is now I would predict corruption in a week - when you use it daily
Regarding the cores - I configured it for 2 cores, and the guest also sees 2 cores (Windows task mgr). What you are saying is very very weird.
Regarding the HD - I already performed shrinking. Will try to convert to pre-allocated and report back.
Alex.
oops - you are right - I misread
numvcpus = 2
cpuid.coresPerSocket = 2
if you had 4 cores it would be
numvcpus = 4
cpuid.coresPerSocket = 2
about shrinking ...
a healthy vmdk should not have any of this entries in a vmware.log
Command READ(10) took 3.173 seconds (ok)
Command READ(10) took 3.246 seconds (ok)
Command READ(10) took 3.206 seconds (ok)
Command READ(10) took 3.618 seconds (ok)
eaither shrinking is long ago - or something else is wrong with your disk-system
Well, converting to a pre-allocated HD solved the problem. Thanks!
Anyway, I believe there is still a problem, as VirtualBox works fine with the same image using a growable VMDK file and VMWare should investigate it.
Alex.
Update
I believe I found the root of the problem. I saw in the log that VMWare complained about libaio. So I installed it, and everything seems to work much better.