VMware Communities
slewsys
Contributor
Contributor

Beta 4 REQUIRES reinstall of XP (32-bit)

VMWare staff recommend reinstalling 32-bit Windows XP under Fusion Beta 4 for improved performance. What they aren't telling us is that the difference is like day and night. VMWare folks seem to concede to those people with dozens of apps installed, that reinstalling may not be an option. I say: UNDER FUSION BETA 4, YOU CANNOT AFFORD NOT TO REINSTALL XP. Even after seeing the recommendations, I resisted since I just don't use XP except as required for clients. But the performance difference is nothing less than stunning. The old (Beta 3) XP VM made the whole machine, both Mac and Windows sides, feel like it was running with leg irons. I attributed this to poor MP support in the Mac kernel. But it is obviously not so simple. XP installed under Beta 4 just flies! There is virtually no performance hit on the Mac side either.

I am running XP Pro (32-bit) under Fusion B4 with debugging disabled, 10GB disk space, not preallocated, 256MB RAM, 1 CPU and 3D acceleration enabled. Running top(1) in a Mac terminal shows an idle XP VM (vmware-vmx) averaging 3.5% CPU utilization on 2.16GHz Core 2 Duo (iMac).

You have been warned!!

Reply
0 Kudos
18 Replies
dkwehe
Contributor
Contributor

Thanks for the note. I'll give it a try in the morning and see if Dragon Naturally is more responsive. These are the types of ideas that are helpful.

Reply
0 Kudos
Rzn8tor
Enthusiast
Enthusiast

Does this mean that you shouldn't use the converter to convert a physical machine to a VM?

Reply
0 Kudos
bgertzfield
Commander
Commander

As long as you end up with a SCSI disk, your performance should be pretty good. That's the key difference between XP 32-bit VMs made with Beta 3 and Beta 4.

Reply
0 Kudos
admin
Immortal
Immortal

And, to Rzn8tor's question: VMware Converter always emits virtual-SCSI VMs, no matter what kind of source VM it takes.

Incidentally, VMware Converter would be a pretty good option for taking a beta3-or-earlier virtual-IDE VM and turning it into a virtual-SCSI VM, if it's too inconvenient to recreate the VM under beta4. The instructions I've posted for converting Parallels VMs (http://www.vmware.com/community/thread.jspa?threadID=88195) would work, and Pat Lee's modification to my technique (http://www.vmware.com/community/thread.jspa?threadID=88195&start=10) would work even better.

Reply
0 Kudos
slewsys
Contributor
Contributor

I am missing something very big here. I have two Windows XP Pro x64 Edition VMs. One that I installed under Beta 3 and, for kicks, another that I installed just now under Beta 4. The Beta 3 VM pulls 20% of the CPU at idle. The Beta 4 VM pulls 4% at idle. And it's not just when the system is idle - there is a huge performance difference - the difference between marginally useful and blazingly fast. What gives? The only difference that I am aware of is that the Beta 3 VM is allocated something like 768MB (out of 2GB) RAM, whereas the Beta 4 VM is allocated 256MB RAM. Everything else is the same, to the best of my knowledge.

My concern is that VMWare is shooting themselves in the foot by suggesting that having a SCSI drive is enough to assure top performance. In my case, that is simply not true. I am happy to provide all of XP VMs to VMWare for them to examine if that would help. In the mean time, I extend my claim: BETA 4 REQUIRES REINSTALL OF XP (32-bit AND 64-bit)

Reply
0 Kudos
admin
Immortal
Immortal

My concern is that VMWare is shooting themselves in

the foot by suggesting that having a SCSI drive is

enough to assure top performance.

We're not suggesting that at all. No physical[/i] SCSI drives are required. Beta4 VMs contain virtual[/i] SCSI drives, while beta3 and earlier VMs contain virtual IDE drives. You can check it out yourself using Device Manager inside the VM.

I understand from my colleagues that our implementation of virtual SCSI I/O has lower virtualization overhead than does our implementation of virtual IDE I/O. However, I'm not an engineer, so don't take that statement as authoritative.

Anyway, like you, I don't understand why that should translate into lower host CPU utilization when idle, as you have seen. But, even before beta4, I was seeing less than 20% utilization anyway. (One possible difference: my VMs are 32-bit, unlike yours, because my Mac has a Core Duo rather than a Core 2 Duo. Poor old me.)

Reply
0 Kudos
tlog
Contributor
Contributor

Id like to add that I followed some guidance elsewhere on these forums, and used a trial version of a disk cloning utility (in my case i used Acronis True Image Home, though others should work fine) and cloned my old IDE disk to a shared folder on my macbook, then restored the image to a new scsi disk that I created in the VM.

I then subsequently deleted the original IDE disk from the VM.

Whole process was painless. Windows did not require reactivation (though MS Office did).

Took 30 mins in total, and performance is much improved.

HTH

Reply
0 Kudos
slewsys
Contributor
Contributor

>> My concern is that VMWare is shooting themselves in

>> the foot by suggesting that having a SCSI drive is

>> enough to assure top performance.

>We're not suggesting that at all. No physical SCSI drives are required. Beta4 VMs contain virtual SCSI >drives, while beta3 and earlier VMs contain virtual IDE drives. You can check it out yourself using Device >Manager inside the VM.

Right, but the Beta 3 VM of XP Pro x64 uses (virtual) SCSI drives, and yet it is still impossibly slow WHEN RUNNING UNDER BETA 4 compared to Beta 4 VM of XP Pro x64. Both use SCSI drives, but they are not equal under Beta 4. This is the point. Given the mixed experiences described on this list, it seems clear that I am not the only one experiencing this. And if my claims hold water, then VMWare needs to clear this up so that people don't walk away never realizing how good it can be ...

Reply
0 Kudos
tlog
Contributor
Contributor

Hmmm Ive done some further (less than scientific admittedly) tests, and puzzled by the results:

http://www.vmware.com/community/message.jspa?messageID=673941#673941

Reply
0 Kudos
dkwehe
Contributor
Contributor

Before I shoot an hour re-installing XP 32 and 64, and all the programs to go with them, does this really make a huge difference? Is there a comparison, say, before and after, or perhaps with the Brand X product?

dkw

Reply
0 Kudos
slewsys
Contributor
Contributor

] Before I shoot an hour re-installing XP 32 and 64, and all the programs to go with them, does this really

] make a huge difference? Is there a comparison, say, before and after, or perhaps with the Brand X

] product?

I suppose it depends on your situation, but the difference between 25% and 4% CPU at idle on a Core 2 Duo is only the beginning when you factor in the difference in disk performance. If you spend more than one hour per week running XP, then you'll quickly recoup the cost of reinstalling with increased productivity.

But you don't have to reinstall all of your programs to verify this performance gain - VMWare Beta 4 makes installing XP (both 32- and 64-bit) basically a one-step process. So while reinstalling your whole system may be a pain, reinstalling just XP itself is trivially easy. Then you can decide for yourself if the increased performance warrants a full reinstall. I'll wager that you'll agree it does. Either way, I hope you'll let us know.

Regarding Brand X, I am not aware of any benchmarks. The only actual figures I've seen are rhind's and the end of the thread: http://www.vmware.com/community/thread.jspa?messageID=673095

Reply
0 Kudos
Pat_Lee
Virtuoso
Virtuoso

] Before I shoot an hour re-installing XP 32 and 64,

and all the programs to go with them, does this

really

] make a huge difference? Is there a comparison, say,

before and after, or perhaps with the Brand X

] product?

I suppose it depends on your situation, but the

difference between 25% and 4% CPU at idle on a Core 2

Duo is only the beginning when you factor in the

difference in disk performance. If you spend more

than one hour per week running XP, then you'll

quickly recoup the cost of reinstalling with

increased productivity.

But you don't have to reinstall all of your programs

to verify this performance gain - VMWare Beta 4 makes

installing XP (both 32- and 64-bit) basically a

one-step process. So while reinstalling your whole

system may be a pain, reinstalling just XP itself is

trivially easy. Then you can decide for yourself if

the increased performance warrants a full reinstall.

I'll wager that you'll agree it does. Either way, I

hope you'll let us know.

Regarding Brand X, I am not aware of any benchmarks.

The only actual figures I've seen are rhind's and

the end of the thread:

http://www.vmware.com/community/thread.jspa?messageID

673095

NOTE: There is NO need to reinstall XP (64-bit) as that guest already used virtual SCSI drives with beta 3, it was only XP that defaulted to IDE.

The advantage of virtual SCSI drives and the improved BIOS boot times combined with defaulting to buffered IO make for a very nice improvement in overall performance for Windows XP 32-bit edition in a VM.

Pat

Reply
0 Kudos
slewsys
Contributor
Contributor

Pat,

There IS a huge difference, even with XP Pro x64. The version installed under Beta 3 still consumes a minimum of 25% CPU WHEN RUN ON Beta 4. An otherwise identical system, only installed under Beta 4, consumes a mere 4% of the CPU at idle. It makes no sense, but that is the pattern of people's experience as described both on this forum and elsewhere. Please try this yourself and let us know the results.

There may be a performance improvement in running a Beta 3 SCSI VM under Beta 4, but that does not negate the huge performance hit compared to installing fresh*. So while technically a reinstall may not be required, I claim that anyone who sees for themselves the difference in performance will consider a reinstall warranted.

Saying otherwise is only hurting VMWare Fusion's reputation because people are judging its performance based on older Beta VMs.

\* I am referring here specifically to XP Pro x64 since I did not try upgrading an XP 32-bit to a (virtual) SCSI disk.

Reply
0 Kudos
Pat_Lee
Virtuoso
Virtuoso

Pat,

There IS a huge difference, even with XP Pro x64.

The version installed under Beta 3 still consumes a

minimum of 25% CPU WHEN RUN ON Beta 4. An otherwise

identical system, only installed under Beta 4,

consumes a mere 4% of the CPU at idle. It makes no

sense, but that is the pattern of people's

experience as described both on this forum and

elsewhere. Please try this yourself and let us know

the results.

There may be a performance improvement in running a

Beta 3 SCSI VM under Beta 4, but that does not negate

the huge performance hit compared to installing

fresh*. So while technically a reinstall may not be

required, I claim that anyone who sees for themselves

the difference in performance will consider a

reinstall warranted.

Saying otherwise is only hurting VMWare Fusion's

reputation because people are judging its performance

based on older Beta VMs.

  • I am referring here specifically to XP Pro x64

since I did not try upgrading an XP 32-bit to a

(virtual) SCSI disk.

I am glad that this helped you a great deal.

Unless this is a Boot Camp VM, I don't know of any changes for x64 in beta 4 that would be causing this difference. One thing with Boot Camp VMs is that we were using a virtual USB mouse for beta 3 vs switching back to a virtual PS/2 mouse in beta 4, which seems to make a big difference. In this case, you don't need a fresh install. You just need to toss the Boot Camp Setting VM and let VMware Fusion recreate it against your existing partition.

I know that my test VMs which were created a while ago and they have been idle performance of 4-8% CPU usage, I didn't make any changes for beta 4. The biggest difference is that I have been using a virtual SCSI disk in my VM for a while.

I think a fresh install in general is a great idea, but other than the 32-bit XP changing from virtual IDE to virtual SCSI in beta 4 I don't want to recommend people for this unless they see a specific need.

Pat

Reply
0 Kudos
dkwehe
Contributor
Contributor

OK, I upgraded to B4.1 and reinstalled XP(32 bit). I reinstalled Dragon Naturally Speaking 9.1. It still runs much slower running on Fusion than on Brand X. The latency is quite noticeable. I'm running on a MacPro with 6GB of memory, so size is not an issue (for this). For my needs, this is the key figure of merit.

I'd be happy to try any suggested tweaks.

dkw.

Reply
0 Kudos
slewsys
Contributor
Contributor

dkwehe: How is your VM installed/configured? You should allocate only 1 CPU (evidently because of shortcomings in Tiger's scheduling, which are hopefully corrected in Leopard) and disable debugging, of course (under VMWare preferences). The only other difference is that, when installing XP (32-bit and 64-bit) on Beta 4, I allocated only 256MB RAM (as recommended) instead of up to 1GB. Had you upgraded your previous VM to SCSI? Was there any improvement in performance between the old VM and the new? I am trying to clarify:

1) Is there really a difference, as I claim, between otherwise identical SCSI-based VMs installed under Beta 3 vs Beta 4 (as I experienced with XP Pro x64)? Or does upgrading, say, an XP VM from IDE to SCSI virtual disk provide the same performance increase as reinstalling?

2) Might there a difference in performance between VMs with different amounts of allocated resources - e.g., RAM (say, 1GB vs 256MB) or virtual disk space (non-preallocated)?

Pat: I have not tried running either XP or XP Pro x64 from a Bootcamp partition. Should there any advantage, performance-wise, to doing so? I am curious what hardware and resources you are using?

Reply
0 Kudos
dkwehe
Contributor
Contributor

Yes, I only allocate 1 cpu, but I allocate 2 GB to Fusion because I need to run Dragon Naturally Speaking. In my comparison, the older VM had not been upgraded from IDE to SCSI. My comparison is beta 3 IDE and beta 4 SCSI both running identical XP installations. I see no difference in my application.

I don't want to sound unappreciative of your wonderful ideas. In fact, progress only occurs because of creative thinkers such as yourself. Please keep your ideas and inquiries coming.

dkwehe

Reply
0 Kudos
slewsys
Contributor
Contributor

dkwehe, thanks for the kind words. A couple more questions: Before running Dragon Naturally Speaking, what is VMWare's CPU utilization, as reported under Mac OS X. For instance, in Terminal.app, I use the command line `top -o cpu -s 10'. With a new XP install, it should less than 5%. Once Dragon Naturally Speaking is running, but otherwise idle, what is the CPU usage then? As discussed on this thread: http://www.vmware.com/community/thread.jspa?threadID=89766&tstart=0 some types of I/O (e.g., iSight) and some applications cause problems. If the system becomes unresponsive only after starting Dragon Naturally Speaking, maybe try another I/O device?

Reply
0 Kudos