Performance dive after upgrading to build 9344

Discussion in 'Installation and Configuration of Parallels Desktop' started by oztrev, Mar 12, 2010.

  1. oztrev

    oztrev Member

    Messages:
    99
    Apple Mac Mini (late 2009)
    2.26GHz CPU
    4GB RAM
    OSX Snow Leopard v10.6.2
    Windows 7 Professional

    Windows 7 Experience Index Scores (old, new build 9344)
    - processor 4.6, 4.5 <<--- loss 0.1
    - memory 4.5, 4.5
    - aero 4.9, 4.8 <<--- loss 0.1
    - 3D 4.1, 3.9 <<--- loss 0.2
    - disk 6.7, 5.9 <<--- loss 0.8

    Ok, so I'm not overly worried about the 0.1 and 0.2 performances losses, but the disk subsystem score dropped by nearly a whole number and the system appears noticeably slower when compiling source code (which was prompted me to re-run the Windows Experience Index score).

    So, is the build 9344 "Improved virtual hard disk caching mode" slowing down the disk subsystem? If so, can it be disabled?
     
  2. mschue

    mschue Bit poster

    Messages:
    7
    Hmm...booting XP takes less than half the time now on my Mini with 1.82GHz CoreDuo, 2GB RAM. I am quite happy with this new release...
     
  3. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Your Windows experience results are interesting... I'm running Parallels on an MBAir 1.6 GHz and this is what I get:

    - processor 4.4
    - memory 3.9
    - aero 2.6
    - 3D 3.6
    - disk 4.2

    Isn't it weird that while you're running a > 2 Ghz processor and I'm running 1.6 Ghz, the processor index is almost the same? While the aero index is quite dramatically different?

    Frankly, I distrust the Windows Experience index. I guess on physical Windows boxes it's a Good Thing (R) but on virtualised ones it is not particularly exact. What's your Real World Experience (R)? Ie do you notice any difference in performance when you look at how fast it runs applications?
     
  4. dev@parallels.com

    dev@parallels.com Parallels Developers

    Messages:
    54
    Win Index

    Windows performance index can't be trusted, really.
    First of all its results deviate and depend on what other activity might wake up during your benchmarking (which doesn't happen on standalone Win host). Next, it can't take into account host OS caching and other such effects.
    So at least remeasure it many-many times and watch host system when taking Windows Performance Index.
     
  5. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    I recently tried the new Fusion 3.1 beta. They're getting there - it's certainly better than 3.0 - but they sure have far to go. However, I read on their FAQ that on an MBAir it actually helps to reduce the number of processors to 1 since the MBAir has a very "aggressive power management policy". I did that on my Parallels VM and things certainly run smoother now (if not quite as fast).
    What I liked about the Fusion beta is that their memory management seems better; even with a 1 Gig VM it was a pleasure to switch between the VM and the Mac. Plus their mouse integration looks a lot nicer (they retain Windows 7's smooth cursor animations). Now if we could have virtualizer with Parallels's unparalleled graphics performance and Fusion's great memory handling...
     
  6. dev@parallels.com

    dev@parallels.com Parallels Developers

    Messages:
    54
    1. Can you point to the FAQ you are talking about? Do you mean reduce number of VM processors (not host), right? Then, it's certainly the case. We do not recommend multiple CPUs as well until you have particular workloads which benefit from additional cores.

    2. "Plus their mouse integration looks a lot nicer (they retain Windows 7's smooth cursor animations)". Actually there should be no difference. Can you described in a bit more details what you mean? Maybe send me an email with screenshots. Would appreciate a lot, thanks!

    3. Regarding the memory manager (MM). To tell the truth MM is 99.9% the same in Fusion and PD. And I was digging multiple cases when people complained about slowdowns or beachballs. And found no product issues. What I really found that Mac MM is really poor and Mac apps have memory leaks or at least their memory usage grows over the time causing swapping caused by VM activity :(

    I have a simple test case on my MBP with 4GB RAM, for example - if you boot VM after MBP reboot it will be fast and no beachballs (even if all your apps opened). If you do the same after a couple of weeks (you can suspend/resume MBP, but not reboot it typically) of working with your apps (like Safari + Mail) you will find out swap used on the host and beachballs on VM start... It's not a VM issue. And it makes it clear that you can compare VM MM only having the same initial conditions. Freshly rebooted host is one of such initial fair conditions.

    My guess is that you started F3.1 VM after the reboot, right?
    Also video RAM assigned to VM must be the same. PD5 allows to change the setting, while F3 - no, it can be checked inside the guest how much VRAM was allocated to it with special tool only.
     
  7. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Edited (added "not" in first sentence)

    Hi dev,

    first - I de-installed Fusion after a lot of testing yesterday night since it's really unusable for anything that's *not* strictly productivity related. I'm on the road a lot, and like to play a game after work every once in a while. And Parallels is unbeaten in graphics performance on my MBAir 1.6 GHz 2 GB Ram.

    1. FAQ: Can't find it anymore. Maybe it was mentioned in the "help" of Fusion. It was very late yesterday. Yes, I mean I reduced the number of virtual processors in the VM, not in the host (I wouldn't know how to turn off one core on my Mac).

    2. Mouse: Windows 7 has very smooth mouse cursor animations (e.g. rotating turqoise ring). When I turn off mouse integration in the Parallels VM, I get the same smooth mouse cursor animations. When I turn on mouse integration in the Parallels VM, the mouse cursor animations get choppy - as if a number of frames had been removed from the animation. Screenshots won't help; I'd need to do a screen movie :) Is the description ok? If not I'll try to make a screen movie.

    3. Memory: Maybe it's not a memory issue - but it sure feels like the Mac starts swapping earlier with Parallels than with VMWare. I had assigned 1 GB to the Fusion VM, and switching between Mac and VM was very fast, and the Mac stayed very responsive. I tried the Fusion VM with 1 and with 2 (virtual) processors - no noticeable difference (3D on, memory caching off). I did get the odd spinning beach ball every once in a while but much less often than when I run a 1 GB Parallels VM.
    The Parallels VM I usually work with has 832 MB assigned to it (3D on, 128 MB video memory), and memory caching is off ("Optimise performance for Mac OS X applications"). This allows me fairly comfortable work in both systems, and both Mac and Windows are quite responsive. But when I up the VM's memory to 1024 MB I run into trouble after a relatively short while, in particular when I have any program running on the Mac. That's why I thought this was a memory handling issue.
    I also noticed that after rebooting the host (the Mac) VM's start up and generally behave much better than after a couple of days without rebooting the host. I shut down my Mac every night (to save current), and in the morning I usually start up the VM first thing (at work. I shower and have breakfast first ;-) - and yes, I ran the VMWare VM after a reboot, but I'm running the Parallels VM also after a reboot.

    Keep up the great work, I love my Parallels VM!
     
    Last edited: Mar 17, 2010
  8. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Hi dev,

    in this context, a question on the Windows swap file. Windows requires some space for its swap drive. However, Parallels *also* generates a memory file, with a .mem suffix (I understand this is to help it wake up faster after a suspend or when the Mac goes to sleep); that memory file contains the current VM memory *and* the Windoes swap file. Why, then, does Windows still need space for its swap file? Is that just a "fake" file to keep Windows happy, or does it actually use that file for swapping? If the latter, wouldn't that mean that swapping is happening twice, once with the Windows swap file, and once with the Parallels memory file?

    I'm just curious.
     
  9. dev@parallels.com

    dev@parallels.com Parallels Developers

    Messages:
    54
    No, there is no double swapping. .mem file is used to keep VM RAM on suspend. While Windows swap file resides in .hdd file (from host perspective) and is used by Windows when it runs out of fast physical RAM (which is in .mem file). Also note, that .mem file though reserves disk space do not perform any I/O during the normal work. It's written to disk only on VM suspend.
    Hope I made it clear. If not - feel free to ask further questions.
     
  10. dev@parallels.com

    dev@parallels.com Parallels Developers

    Messages:
    54
    Yeah, looks like I see what you mean... Thanks a lot for the report, we will take a look why is that.
    And thanks a lot for your kind words :)
     
  11. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Hi dev,

    thanks, your description of the .mem file and its function makes matters clearer to me. One question remains, though: I took a look at the .mem file while the corresponding VM was running by using cmd-i on the file to bring up the Mac's information dialog on the file. I found that the file was updated every few minutes. If it isn't being written to while the VM is running, why is it updating every few minutes? Doesn't that cost performance?
     
  12. dev@parallels.com

    dev@parallels.com Parallels Developers

    Messages:
    54
    No, it's just modification time of the file changes on some minor operations. But real content is locked in memory and host OS does nothing with it.
     
  13. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    OK, thanks for the heads up :)
     
  14. Doug Kolb

    Doug Kolb Bit poster

    Messages:
    9
    (MBP - 6GB - Firewire 800 disks - Build 9344) After I launch Parallels it takes FOREVER to do anything else. It sometimes happens immediately, and (almost?) ALWAYS happens soon after launching a substantial Mac app such as Photoshop. I am having the problem with 9344 and any version of Windows, although I use Win7 most often. Using Mac Activity Monitor, I have observed that Parallels does not seem to release memory, even after you quit the app. As a result, the only way to regain performance is to restart the system. I read through all of the posts above. None of the the posts seem to address this aspect of the problem, yet it seems to me that this GIANT memory link just might be the real cause of the problem. To duplicate, all you need to do is run IE in a VM and watch memory usage during launch and after quitting Parallels.
     
  15. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Hi Doug,

    hm, I can't confirm that - once I quit Parallels (and that means the app, not just shutting down the VM) I get all the memory back, at least according to Activity Monitor (and this is on a memory-challenged machine, a 2 GB MBAir). Try running your VM with 1 GB of memory for a while and see if this phenomenon persists.
     
  16. Darkanima

    Darkanima Bit poster

    Messages:
    3
    Slow shutdown

    I still have problems with build 9344 slow shutdown. I've done a few tests and these are the shutdown times:

    Macbook Pro without paralells or vmware fusion: ~2 secs.
    Macbook Pro with VMware fusion installed: ~2 secs.
    Macbook Pro with Parallels installed (launched or not): ~15 secs.

    My system:

    Model: MacBookPro5,3
    Processor: Intel Core 2 Duo@2,8 GHz
    Memory: 4 GB
     
  17. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Darkanima,

    could you post your VM specs - how much memory, how much video memory, and how many processors have you assigned to your VM; which OS are you running (XP? Windows 7?); and are you running it in coherence / crystal, full screen, or windowed view?
     
  18. Darkanima

    Darkanima Bit poster

    Messages:
    3
    Well, I've been doing some additional tests:

    It doesn't matter how much memory, cores,... I assign to the VM, it always takes the same amount of time to shutdown the whole system.

    My VM settings are two, depending on the task, always in full screen(4 space out of 4) with Windows 7 Ultimate and optimized for VM:

    If I need performance for games: 2 cores, 1256 MB Ram, 256 MB of Video memory.
    If I'm not going to play only: 1 Core, 1024 MB of Ram, 128 MB Video memory.

    I've been measuring several times the overall shutdown of the system, and as I said before it doesn't matter if, in that session, I started Parallels or not, or if I do I quick restart after the start without launching any app, It always takes the same time:

    ~8 secs until I see the blue Mac OS screen before shutdown and ~2 secs until it is completely shutdown.
     
    Last edited: Mar 24, 2010
  19. Shaddam IV

    Shaddam IV Forum Maven

    Messages:
    623
    Optimize for OS X and test again. But yes, after I run the VM extensively I have similar shutdown times (which to me, subjectively, are about twice as long than shutting down the Mac without having used the VM). I guess the Mac is cleaning up its virtual memory before shutting down. I hear from the Parallels developers (but not only them) that the Mac OS's virtual memory handling leaves room for improvement, to put it politely.
     
  20. Darkanima

    Darkanima Bit poster

    Messages:
    3
    I've tested using optimize for OS X applications, but it still takes the same time.
     

Share This Page