I should add that vmtool is installed (after XP is installed as the guest OS), but the heavy disk usage occurs DURING the XP installation.
Have you tried disabling the vRAM caching (Swapping) of the vm's? Via VS Console under host properties - make sure the vm's only use the host memory and disable swapping of mem.
What is reported in the vmware.log file from the virtual machine?
I just tried disabling swapping on the guest OS. The boot-up time is improved (about 2.5 minutes). However, the guest OS (xp) is still not usable. Things like clicking on "My Computer" would still cause heavy disk-usage.
It is as if no physical RAM is available to the guest OS, and as such it is using the disk as the RAM.
I have the log available but it is pretty long and I don't want to post it here. Is there anything in particular I should be looking for?
I do notice there are a lot of entries relating to the HD usage:
Jan 10 21:34:18: vmx| DISK: DISK/CDROM timeout of 2.493 seconds on ide1:0 (ok)
Jan 10 21:40:05: vmx| DISK: DISK/CDROM timeout of 1.692 seconds on ide0:0 (ok)
Jan 10 21:44:08: vmx| DISK: DISK/CDROM timeout of 4.024 seconds on ide0:0 (ok)
Jan 10 21:45:02: vmx| DISK: DISK/CDROM timeout of 1.284 seconds on ide0:0 (ok)
Jan 10 21:46:26: vmx| DISK: DISK/CDROM timeout of 1.240 seconds on ide0:0 (ok)
and it goes on. Would this be the cause of the heavy disk usage?
For virtual machines with large ammount of memory, disabling the memory trimming fixed some of the slow disk access issues.
One other good thing that fixed my slow disk access issues was to disable the write cache on each virtual hard drives I had.
Hope this helps.
mainMem.useNamedFile = FALSE
worked well for me when I was running VMs with 3GB of memory on VMware Server.
I had very similar problems with Open SuSE 10.0 (all patches applied) running on a 2 GHz Centrino (single-core)-based laptop with 1 GB RAM. The VM was Windows 2003 Server with 512 MB. Performance was so sluggish as to make the VM completely unusable
it was odd, thoughcertain things, like tracking the movement of the mouse was pretty quick and responsive, but other things like clicking a menu choice or clicking an item to start a program or performing an action in a program on the Windows VM took forever. Even so simple an act as starting up or shutting down the VM took tens of minutes.
I resolved the problem by dumping SuSE in favor of CentOS 5 (Also runs well under CentOS 4,).
I expect that the problem may have to do with the way the kernel was compiled for SuSE. I don't know a whole lot about it, but I understand that some of the newer kernels are configured to interrupt 1000 times per second to update internal chronometers (system clock, etc.) while older ones interrupt only 100 times per second. I have also seen kernel options (but, alas, do not remember what they are) that allow for different levels of granularity in switching between tasks. Again, the older method is highly preemptive and focuses more or less on server performance, while the newer methods seemed to be focused on making the way the multitasking controller code switches task more finely grained and more biased toward giving the user's desktop a "more responsive/faster" feel.
Unfortunately, these new optimizations for 1000 times/second timer interrupts (or whatever) and/or the new multitasking performance "tweaks" than focus on enhancing the desktop experience are anathema for VMware Server.
Symptoms that I have noted include, but are not limited to:
Very slow VMware VM performance (as detailed above) accompanied by abnormally LOW cpu utilization on the host.
Possible very high disk activity (perhaps high "load" values as reported by TOP or similar utilities)
Host operating system will seem highly responsive and quick--just the VM(s) running under it seem to run in "slow motion".
The easiest cure was to dump the desktop distribution for a "server-oriented one", such as SuSE Linux Enterprise Server or RedHat/CentOS Enterprise Linux, which will have a kernel optimized for performance of server tasks (or, especially virtual machines).
An alternative is to recompile your kernel with the server-type optimizations turned on and the newer desktop performance-oriented ones turned off. You may also wish to attend to changing the timer interrupt value mentioned above from 1000 to 100.
You will have to research the compile options to find out how and where to set these parameters in your kernel Makefile as, unfortunately, I do not recall them.
A third option is to try a kernel of the 2.6.21 or later series. I have heard of various problems with kernels in the 2.6.15 through 2.16.19/20 range. (I have a Gentoo-based laptop with Kernel 126.96.36.199 (or something like that) and this machine runs VMware Server exceptionally well. However, there may be some issues with getting VMware to work with such a new kernel, e.g. you may need to apply something called the "any-to-any" patch to VMware Server to get it to compile and run properly.
Like I said: Dumping SuSE for some variant of RedHat or CentOS Enterprise Linux is probably your easiest and most reliable solution.
I too notice bad performance.
2 XP Prof Workstations with 400 MB RAM each need some minutes for logon. They are running on a HP Server with 4GB RAM, Server 2003
After a reboot of the VMs everything is well for a short time.
after reboot each VMWare process gains the full 400 MB.
After some time this is much less - even after setting "Disable memory page trimming".
VMWare Server Host Setting is "Fit all machine memory into reserved RAM" - reserved RAM ist 1545 MB
VERY noticeable: I checked RAM consumption: it was 300 and 200 MB for the 2 VMs.
I then contacted one VM with mstsc.exe. Login again was very slow. I checked memory consumption in Task Manager: it is now actually
110 MB and 17MB (!!!!!)
Of course performance is very bad.
How can I prevent that? I cannot boot the VMs all the time to get some performance...
Message was edited by:
Because you're on Windows you should rather open the new thread.
What storage system is used on the host? What other programs are installed?
2 Disks mirrored for OS (Server 2003 SP1)
3 Disks Raid 5 for Data storage (450 GB) - here are the 2 VMs
CPU: 2 x XEON processor 3 GHZ
RAM: 4 GB
VMWare Server program version is 1.0.2 build 39867
Server ist file- and printserver and has "nothing to do". cpu usage is 2 to 10 %.
I just rebootet the XP Prof VM - and I have normal performance.
Task manager shows for this process memory usage of 379MB.
This works fine until the memory for this process is "stolen"
One question: after setting the advanced VM setting "Disable page trimming": should I restart the server?
Ouch - RAID5 isn't the best if you want the best VM performance. What type of RAID controller do you have? It can work okay if you have a decent (read: expensive) RAID controller, but the cheaper RAID controllers just can't keep up with the striping and parity needed for RAID5 and maintain a good level of throughput.
Its the standard Raid 5 controller of a HP ML 350 server.
Its not the key.
Notice: I have good performance with the 2 XP Prof VMs when task manager of the server shows the 400MB RAM for each VM process.
This becomes less with the time - then the VM processes show only about 100 MB of RAM in task manager.
Performance then is very poor!
I reboot the VMs - task manager shows 400 MB RAM for each VM process. Now performance is good again.
Performance should increase as utilization of the VMs increases. When you leave a VM idle, VMware gradually releases physical memory back to the O/S for other tasks. If you start using the VM again, VMware should take that memory back and reallocate it to the VMware process, unless, of course, you have something else running on the host that won't allow that memory to be reallocated.