Fusion Beta 3 vs Parallels performance

[manuel_wenger]

I can consider myself a VMware fan, as we're using several ESX servers at our company, and I've always used VMware Workstation for Windows. Before Fusion Beta 3, I've always used Parallels to virtualize Windows on my iMac at home.

Now I wanted to switch to Fusion, so I migrated my existing Parallels (Build 3188) VM to Fusion succesfully. Everything works.

My only issue is performance: even with debugging disabled, Fusion seems to be quite a bit slower than Parallels. I'm not talking about plain computing power here (Fusion seems to be faster than Parallels doing calculations even according to the Parallels forum members) but about "apparent sluggishness" and application startup/reaction times.

I have the same settings on Fusion and Parallels (256 MB RAM allocated for a pretty simple WinXP Pro VM with almost nothing loaded). But when I start up my average application (for example Netobjects Fusion or the Sony Aibo Entertainment Player) which takes, say, 5 seconds to load under Parallels, it will take 10 seconds under Fusion. During the additional time it takes, it's apparently doing nothing - not even using 100% CPU power or something.

It's not that typical "everything is swapping to disk"-sluggishness - it's definitely not swapping. It must be something else. Tools are updated.

What else could I be missing? Everyone seems to say that Fusion is faster than Parallels, but on my system it's quite a bit slower. iMac 20 2.16GHz C2D, 2GB RAM, OS X 10.4.9.

Could it be the fact that it's a migrated VM, therefore not ACPI? Or is this irrelevant?

Thanks

-Manuel

[SSonnentag]

Make sure you have debug mode turned off in the options. That's the only thing I can think of off hand.

[admin]

Because you're seeing the slowness when starting up programs, it sounds like a disk access problem. Try flipping the hidden aiomgr.buffered switch (equivalent to Parallels' "Guest OS Performance Optimized") as described in beta3 speed .

[manuel_wenger]

I had already turned off debug mode, I just forgot to state it in my original post.

But that aiomgr.buffered switch definitely made the difference! Now Fusion \*is* faster than Parallels ... way to go VMware!

Thanks etung for the tip, I had completely missed that thread for some reason.

[donmontalvo]

this should be a sticky or part of an faq. some of us are just getting started with vmware (or trying to, still wrestling with the bsod ). at some point users will need to glance at the major issues/solutions quickly. stickies/faq would make migration better for everyone.

don

[Linh_My]

Thank you. Now perhaps someone from VMware can tell us why this switch and the sound switch are so effectively hidden, at least in Win2k. I could understand it from an enterprise aspect, well sort of. From a consumer product hiding these things doesn't make much sense to me.

[Pat_Lee]

Regarding the question as to why the sound and buffered IO flags are hidden, it honestly comes down to our intent with VMware Fusion. Since we are designing a Mac consumer product, we need to make sure to just do the the right thing out of the box and have configuration options only when they make sense. Since we are still in beta, we haven't nailed it correctly in all cases. As we get closer to shipping, we will resolve these issues and make sure the defaults are right and it just works as expected out the box.

Our near term goal is posting select VMX settings in the forum to either work around issues we didn't catch before we shipped beta 3 and to provide workarounds or changes as quickly as possible before our next release.

We will continue to look at user feedback to help verify we are setting the correct defaults or not and to make better appropriate choices to give our users the best possible experience when we ship.

Best,

Pat Lee

Senior Product Manager - Mac Products

VMware

[aliasme]

A couple of observations relative to options:

1) The option dialog box is very hard to work with. There has to be a better way than trapping the mouse.

2) When setting up an OS, users should have access to some more advanced options without canceling the workflow. For example, I was setting up Windows and knew I wanted dual CPU and more memory. I also wanted to use the automated setup, but when I chose not to continue the setup to adjust cpu/mem I also couldn't get the automated setup to continue. When launched, the setup was a normal boot.

[Piggy]

Regarding the question as to why the sound and buffered IO flags are hidden, it honestly comes down to our intent with VMware Fusion. Since we are designing a Mac consumer product, we need to make sure to just do the the right thing out of the box and have configuration options only when they make sense.[/i]

Pat,

Here's my two cents on this issue in response to someone that requested hidden, advanced features be made available.

I'm a 20+ year IT veteran. I'm a professional services consultant for a software company. My job is to deploy our software which runs on Windows and several unix flavors. There are some folks in sales and sales engineering that use Macs but I'm the only PS consultant that does. Our IT department doesn't support Macs so I keep a low profile and don't use the PC they gave me which includes VMware.

I use VMware instead of Parallels (which I also own) because IMHO it's the IT professional's choice and regardless of how it compares to Parallels, I expect VMware to be the professional choice.

Last year when I would show up at a customer with my Mac they were intrigued. Now, I rarely get a second look. Last month I was onsite working with an Oracle DBA hat mentioned he was thinking about getting a Mac. A few days later he had one on his desk. I've stopped counting the number of customer IT professionals that tell me they recently bought one. Apple is tapping an entirely new market. Not just Windows users, but IT professionals who historically have been the first ones to say 'no' to the corporation supporting Macs.

I work with a lot of professionals and they vary in skill sets and professional training however just about everyone that manages dozens (not to mention hundreds or thousands) of Windows and/or Unix servers knows how to do some performance tuning. We really do need access and documentation on more advanced features because we can really benefit from being able to tweak performance for different needs. I myself run Windows XP with the most current version of our software yet also have XP VMs with previous versions and am currently building a Wndows 2003 server VM. I want performance of OS/X to remain snappy while running 1-2 VMs. I want it all!

On the subject of performance tuning, I would like to see an automatic tuning feature for each VM. For example, should I set the swap file to a fixed size or let Windows resize it as needed. I'd like VMware to monitor this and make recommended settings. Should I disable some of the cosmetic, visual Windows features? Personally I'd prefer to leave them enabled but I'd like to know what the performance effect is of each one so that I can make some decisions. VMware should help with this. Which OS is a bigger resource hog - XP or Vista? I'd like to know because I can use either for my purpose but assume XP is the better choice.

Well that's my two cents, or maybe that was $2. Thanks for listening.

[pjv]

wow, thanks a lot etung (and bgertzfield). for me, this made a truly large difference in the perceived performance of my guest (win2k) vm. not just booting or starting up programs either. the whole guest (clicks, drags, window minimizing, etc) now snaps around like a native install. for someone who sits in front of the box all day, the contrast with the previous sluggishness really makes a big difference.

vmware folks: can i suggest that you create some additional preferences for fusion, maybe labeling them 'advanced' or something if you think most people should not monkey with them, and include this option in there?

and given the extreme difference that this option made for me in my subjective (read: what matters if i am deciding whether to buy software or not) experience with fusion, i highly recommend that you consider enabling this option by default for users who have some threshold amount of physical ram. and i would make this threshold pretty low. i only have 1.5 gb in my box, with 628 mb allocated to the guest, and enabling this option made a tremendous difference in my vm's performance while not causing (at least so far) any significant slowdown in my overall host performance.

numeric benchmarks are fine for brochures, but if you want to sell it to people who are demo-ing it (and especially comparing it with another, functionally equivalent app), then tune the app for maximum snap.