Subscribe via RSS
administration c# cocoa coding EasyTAG linux Mac OS X opensolaris OpenVPN Time Machine virtualization windows xcode zfs

Parallels 7 vs Virtualbox 4.1 vs VMWare Fusion 4.1

 

I recently needed to decide which virtualization software to use on my new MacMini.

I see three main competitors for Desktop Virtualization on the Mac:

  •     Paralles Desktop for Mac (~ 80 €)
  •     VMWare Fusion (~ 70 €, now 40 € introduction price)
  •     Virtualbox (OpenSource / with Extension pack usage under their PUEL License)

Parallels and VMWare are commercial products while Virtualbox (at least the main part) is now OpenSource and if you use their Oracle Expansion Pack is at least free for personal use under their PUEL License).
It proved to be hard to find some serious comparison of the three at a recent version. So I decided to make my own. Fortunately VMWare and Parallels provide time-limited demos of their software for evaluation purposes. I took that offer. Installed the three and made some testing and benchmarks.

The tests were conducted on a MacMini 2011 with 2,3 GHz Core i5 Processor and 8 GB RAM and the Stock 500GB Toshiba MK5065GSFX HardDrive.

I installed each of the three on a fresh installation of Mac OS X (10.7.2), rebooted between startup tests and bechmarks to avoid caching effects. And waited at min 1 Minute after Startup to ensure all system services and startup items were done.
No additional software was installed on the machine, just bare Mac OS + one of the virtualization software products.

Tested product versions were : Virtualbox 4.1.6, VMWare Fusion 4.1.1 and Parallels 7.0.15052 . Each virtual machine was installed with standard options, given 2GB of RAM and 20 GB of dynamically expanding harddisk.

Windows startup performance was measured with a stopwatch, taken 7 times, best and worst time were thrown away and the other times averaged for each product. Taken was the time from pressing the "Start VM" button until the Desktop background first appeared, this means these times include startup and initialization of the VMs and it's bios.

 

Results

  Parallels Virtualbox VMWare Fusion
Windows 7 Startup 36 44 39
Crystal Disk Mark 4k Write 0,94 MB/s 0,91 MB/s 1,38 MB/s
Geekbench 3960 3919 3926
PCMark 7 – Overall Score 2007 778 2042
PCMark 7 – Video playback 13,77 fps 7,69 fps 13.76 fps
PCMark 7 – DirectX 9 9,96 fps 1 fps 11,19 fps
PCMark 7 – Data decrypting 12,68 MB/s 10,85 MB/s 17,77 MB/s
PCMark 7 – Video transcoding downscaling 706,35 kB/s 543,93 kB/s 679,09 kB/s
PCMark 7 – Image Manipulation 7,07 Mpx/s 6,93 Mpx/s 7,05 Mpx/s

Analysis and remarks:

In Windows 7 Startup performance the clear winner in my tests was Parallels Desktop. However I assume they use some startup optimization by modifying the windows boot process. How come ? Well Virtualbox and VMWare show the normal boot screen just like a regular PC for Windows 7 ( the three moving colory things), while Parallels shows some Vista style progress bar. Hence I assume they somehow change and enhance the bootup process to get this clear advantage.
I could probably have tested some Linux / Mac operating system startup performance, but that would have increased my test time more than I would have wanted – maybe next time.

Unfortunately I had to disable 3D acceleration for the Virtualbox tests, when it was turned on PCMark would crash for me right after starting. This leads to bad graphical performance for Virtualbox in DirectX, Video Playback and other graphical parts of PCMark and thus results in an overall "bad" PCMark Score not directly comparable to the other two.

Crystal Disk Mark showed odd high values for Parallels and Fusion read tests, most likely because they use some form of caching, which is why I only included 4k write values here. They show a huge (~ 50%) advantage for Vmware.

Geekbench is a computational intensive cross platform benchmark, that stresses mainly the CPU with Encryption, Compression, Image Manipulation and mathematical tests. All products are in a very close range here, with a variance of approx 1%. You could say that all three products perform almost identical.

 

Conclusions :

In Windows startup performance (at least for Windows 7 32bit) the crown clearly goes to Parallels Desktop for Mac. However I did not test VM suspend/resume performance thoroughly which is probably at least as interesting …

Pure CPU computing performance (geekbench, PCMark image manipulation etc.) just differs by about 1% between the three tested products. So if you run just CPU intensive stuff like whatever (image manipulation, scientific calculations) that does not use much data from the disk it does not matter much, which one you choose. Even though I would attest  Parallels a quantum of advance in this field.

Putting caching effects aside, VMWare seems to achieve the best disk IO performance.

In my tests VMWare also achieved the best 3D graphcis performance. Here you have to consider that Virtualbox is not as bad as 1 fps suggests, I have successfully tested games using Virtualbox on my older macbook, so in general 3D works there too, it just did not work with PCMark 7 in my tests. However I think Fusion does best when it comes to 3D Applications.

 

My Verdict :

Parallels Desktop for Mac
+ choose if you want absolute fastest CPU performance, even though only by a margin
+ very good Mac integration (copy & paste between Mac Desktop and Windows VM)
– faster release cycle with a lot of paid upgrades
– highest Retail price

VMWare Fusion
+ best IO performance
+ best 3D performance
+ very good Mac integration (copy & paste between Mac Desktop and Windows VM)
+ can be used with one license on a Desktop and a Laptop by one person (> link)
– offered no upgrade between Fusion 3 and 4 at all.

Virtualbox
+ Free – obviously the best price
+ OpenSource (at least the main part)
– worst IO / 3D performance of the three

 

If you only occasionally use a virtual machine or two. I would choose Virtualbox. Free is simply too good of a price. It works good, has almost all the features the other two have – but unfortunately doesn't have such good Mac integration as the other two.
I choose Fusion – why ? It has good performance, all the features I need and I can install it on my MacMini and my Macbook simultaneosly, so I have my VMs when I'm on the road.
I once used Parallels, but I didn't like their paid upgrades which came quite often ( and which I would have needed to run it on Snow Leopard). So I cross-switched to Fusion in a promotion for about 10€. I'm quite happy with it. VMWare has a huge userbase with all their virtualization products across all platforms and you can google solutions to almost everything related quickly. Unfortunately they do not provide an upgrade at all from 3 to 4, however the posibility to run it on all my Macs makes that up for me. So I took Fusion 4 with the promotional discount for about 40 €.
 

9 Responses to “Parallels 7 vs Virtualbox 4.1 vs VMWare Fusion 4.1”

  • I’d like to see this IO benchmark performed using raw partitions for the guest VMs, for it to be virtual-disk-format agnostic.

  • Hello,

    I have also evaluated VMWare Fusion and VirtualBox… and have seen the same behaviour.
    VirtualBox does work and consumes considerably less RAM than VMWare… but it is much, much slower during disc operations… the installation itself took three times longer time than VMWare.

    My opinion is that VMWare Fusion is worth that amount if you use the VM on every day basis.

    Regards,
    Z

  • What other plans do you have for this system? I mean, are you planning to run the free ESXi Hypervisor 5.0 or something else on it? What is the budget you’re looking at for this home system? Recently I’ve ordered several HP ProLiant ML110 G7 servers to run a single virtual machine within the free ESXi (both Windows 7 Professional and Windows Server 2008 R2) and have had wonderful luck with them. Granted, in all of them I added an HP SmartArray P410/256 MB RAID controller to enable full support for RAID arrays in the ESXi, but it has worked very well, and the performance has been great. These servers are quite affordable when you look at all the features you get with them, and from what I can see they have all your virtualization features with the Sandy Bridge Xeon processors. Your only limitation with the ML110 G7 server is the maximum memory is 16 GB, but that’s still quite a bit to throw at a single VM and well within the range to use the free ESXi Hypervisor.

  •  

    I have no proven results on this matter, but it should be possible to measure the products on a performance per Watt basis. 
    You can of course measure the cpu/gpu time taken to perform a specific operation inside the VMs on different Host Software and put in in relation to the performance produced in the VM, to deduct the efficiency / overhead of the Virtualization Software.
     
    However this would increase the complexity of testing significantly and be much more complicated than my measurements.
     
    If you are lucky you can find such results somewhere, but I highly doubt that and if you find such results you would be very lucky if they compare current versions of the Virtualization software – that’s why I made my own measurements …
     
    Regards Marco
  • Is there any way to compare the 3 in terms of system usage?
    How much of the (real) CPU, HDD,GPU,… is used to operate the VM system.
    The reason i am interested in this is because i’m would like to choose 1 of these 3 products to run on my Macbook Pro and a very important factor for me is battery life.

    do you have any insight into this?

    Anyway, a very good review. thanks for taking all the time.

  • Thanks for your time and effort. Also thanks for sharing, it was a big help and the most comprehensive comparison i could find!

  • It looks like it might help to make a decision to a lot of people. Thank you Marco!

  • I’m glad it helped a bit 🙂

  • Was debating if I should fork out for VMWare, or just go for the free option. Now feel a little happier to spend the 50 bucks for VMWare.

    Thanks for taking the time to post the results of your testing.



Your comment :
Name *

Email*
(will not be published)

Website