Core ix Virtualization Benchmarks?

[savar]

One of the big reasons why I'm holding out for a core ix laptop is that it should have better virtualization performance than the existing C2D laptops.

But I'm beginning to wonder... how much better? Does anybody have benchmarks comparing virtualization performance between the ix and the C2D?

On google I could only find a few benchmarks that compare the i7 virtualization against a native host... WTF is the point of that? I doubt many people are out there trying to decide whether to virtualize or run on bare metal.

Anyway, I'm beginning to think about just buying a used C2D and being done with it. My obsession with Arrandale is just a huge distraction and my current laptop is an everyday annoyance.

On the other hand, if the core ix has 10% better virutalization performance, than that changes the equation.

Edit: This is the article I was referring to: http://www.phoronix.com/scan.php?page=article&item=intel_corei7_virt&num=2. Pointless measurements, IMO.

 

[snaky69]

One of the big reasons why I'm holding out for a core ix laptop is that it should have better virtualization performance than the existing C2D laptops.

But I'm beginning to wonder... how much better? Does anybody have benchmarks comparing virtualization performance between the ix and the C2D?

On google I could only find a few benchmarks that compare the i7 virtualization against a native host... WTF is the point of that? I doubt many people are out there trying to decide whether to virtualize or run on bare metal.

Anyway, I'm beginning to think about just buying a used C2D and being done with it. My obsession with Arrandale is just a huge distraction and my current laptop is an everyday annoyance.

On the other hand, if the core ix has 10% better virutalization performance, than that changes the equation.

Edit: This is the article I was referring to: http://www.phoronix.com/scan.php?page=article&item=intel_corei7_virt&num=2. Pointless measurements, IMO.

IMO the real bottleneck for virtualization performance is the amount of RAM you can allocated to your virtual machine. I don't think any more processing power will change much.

 

[savar]

IMO the real bottleneck for virtualization performance is the amount of RAM you can allocated to your virtual machine. I don't think any more processing power will change much.

That's only a bottleneck if the guest OS + VMWare needs more memory than you have available. But a linux guest with no X11 can run in 512 MB comfortable, and VMWare only adds a few hundred MB on top of that at worst.

That's not really the answer I'm looking for.

What I'm actually interested in is whether the improved VT-x and VT-d technologies in Nehalem are

a) Enabled for the Arrandale CPUs
b) Make a difference in real-world virtualization performance.

 

[savar]

Well...great. This thread is now the #1 hit on Google for for "core ix virtualization". So there is officially NO information about this anywhere in the tubes.

It's disheartening to be on a forum full of nerds and ask the one question that even the other nerds don't want to talk about

 

[Gen]

Well...great. This thread is now the #1 hit on Google for for "core ix virtualization". So there is officially NO information about this anywhere in the tubes.

It's disheartening to be on a forum full of nerds and ask the one question that even the other nerds don't want to talk about

Go out and borrow a windows laptop with any iX core - run your virtualization tests, then run the same tests on a C2D machine - and let US know your results.

Simple.

 

[gwsat]

IMO the real bottleneck for virtualization performance is the amount of RAM you can allocated to your virtual machine. I don't think any more processing power will change much.

That's only a bottleneck if the guest OS + VMWare needs more memory than you have available. But a linux guest with no X11 can run in 512 MB comfortable, and VMWare only adds a few hundred MB on top of that at worst.

That's not really the answer I'm looking for.

What I'm actually interested in is whether the improved VT-x and VT-d technologies in Nehalem are

a) Enabled for the Arrandale CPUs
b) Make a difference in real-world virtualization performance.

Well...great. This thread is now the #1 hit on Google for for "core ix virtualization". So there is officially NO information about this anywhere in the tubes.

It's disheartening to be on a forum full of nerds and ask the one question that even the other nerds don't want to talk about

Nah, in the way of all nerds, we don't want to talk about Core ix virtualization benchmarks because we don't know anything about them and are afraid of proving it.

More seriously, I have only run Windows in VMware Fusion virtual machines, so can't comment on the memory requirements of a Fusion partition when it's hosting Unix apps. In my case, though, having only 2Gb of RAM in my MBP brought Fusion and Windows to their knees. But when I upped the MBP's RAM to 6Gb, with 3Gb of it dedicated to fusion, both the speed and stability of Windows apps improved dramatically. I was, for the first time, able to run them In Unity mode from the OS X desktop. Thus, I have proved to myself at least that Fusion running Windows is a real memory hog.

 

[savar]

In my case, though, having only 2Gb of RAM in my MBP brought Fusion and Windows to their knees. But when I upped the MBP's RAM to 6Gb, with 3Gb of it dedicated to fusion, both the speed and stability of Windows apps improved dramatically. I was, for the first time, able to run them In Unity mode from the OS X desktop. Thus, I have proved to myself at least that Fusion running Windows is a real memory hog.

That's good to hear. That's my main reservation right now. I *can* run Win7 in a virtual machine, but its painful and I try to avoid it.

Go out and borrow a windows laptop with any iX core - run your virtualization tests, then run the same tests on a C2D machine - and let US know your results.

Simple.

I wish! I don't know anybody who has an iX. I have a C2D Dell at work, and a CD MacBook at home. I was going to try to run the VMmark benchmark, but it turns out its really complicated to set up! I don't think it was ever designed to be run on a laptop -- its for hardware vendors that want benchmarks to brag about, i.e. Dell 16 core Xeon monsters.

But -- at last -- I found some documentation on the virtualization performance advantage of Nehalem.

Nehalem's main contribution to virtualiztion: Extended Page Tables (EPT) and Virtualized I/O. I don't think that the latter really applies to desktop virtualization, but EPT probably does.

VMware published a paper on the performance advantage of EPT.

It gets pretty technical, but the basic result is that for MMU intensive benchmarks (their benchmark is compiling the Apache web server from scratch), they saw as much as 48% gain with EPT. On other benchmarks (they use an Oracle benchmark that is not MMU-intensive), there is no significant improvement.

Bottom line: EPT probably doesn't mean much for desktop virtualization, and therefore Arrandale is not a game changer for those of running Linux or Windows in a VM on our MBPs.

 

[gwsat]

That's good to hear. That's my main reservation right now. I *can* run Win7 in a virtual machine, but its painful and I try to avoid it.

I am convinced that the speed and stability of Windows apps running in Fusion on a Macbook Pro is mostly a function of how much memory you have. By current standards, my Macbook Pro's 2.4Ghz Core 2 Duo CPU is hardly the most capable kid on the block. Nevertheless, with 6Gb of RAM to work with, I could hardly ask for more when running Windows apps.

 

[ayeying]

I am convinced that the speed and stability of Windows apps running in Fusion on a Macbook Pro is mostly a function of how much memory you have. By current standards, my Macbook Pro's 2.4Ghz Core 2 Duo CPU is hardly the most capable kid on the block. Nevertheless, with 6Gb of RAM to work with, I could hardly ask for more when running Windows apps.

It depends on how you use it, what type of hard drive you have also.

Running Windows XP with 3GB out of 6GB ram on the iMac just seems sluggish. It's more sluggish then me running Windows 7 on my MacBook Pro with only 2GB ram and only 768MB dedicated. It's entirely smooth and I leave it constantly.

 

[savar]

It depends on how you use it, what type of hard drive you have also.

Running Windows XP with 3GB out of 6GB ram on the iMac just seems sluggish. It's more sluggish then me running Windows 7 on my MacBook Pro with only 2GB ram and only 768MB dedicated. It's entirely smooth and I leave it constantly.

Which iMac? A core duo or core 2?

 

[gwsat]

It depends on how you use it, what type of hard drive you have also.

Running Windows XP with 3GB out of 6GB ram on the iMac just seems sluggish. It's more sluggish then me running Windows 7 on my MacBook Pro with only 2GB ram and only 768MB dedicated. It's entirely smooth and I leave it constantly.

What virtualization program are you using? My remarks were based on running Windows XP and Fusion 2 with only 1Gb of Ram and 1 processor core dedicated to Fusion. Opening up XP again after Fusion had suspended it took forever. If I left Fusion and Windows open, when I would go back to Windows, it would stay frozen for several minutes before I could use any Windows app. Thus, I didn't even try Unity mode.

After I upgraded to 6Gb of RAM I also upgraded to Windows 7 and Fusion 3. with 3Gb of RAM and 1 processor core dedicated to Fusion, Windows opens in less than a minute and I routinely leave it and Fusion running in Unity mode so that I can use Windows apps from the OS X desktop. I didn't even try to do that when my MBP had only 2Gb of RAM. Even now, Windows apps are not quite as fast as OS X apps but that was to be expected of any program running in a virtual environment. Nevertheless, the speed of Windows and its apps has been satisfactory.

My hard drive is the 160Gb Hitachi that was original equipment on my 2007 Macbook Pro.

 

[ayeying]

Which iMac? A core duo or core 2?

It's in my sig. It's a 2.4GHz Core 2 Duo T7700 Processor.

What virtualization program are you using? My remarks were based on running Windows XP and Fusion 2 with only 1Gb of Ram and 1 processor core dedicated to Fusion. Opening up XP again after Fusion had suspended it took forever. If I left Fusion and Windows open, when I would go back to Windows, it would stay frozen for several minutes before I could use any Windows app. Thus, I didn't even try Unity mode.

After I upgraded to 6Gb of RAM I also upgraded to Windows 7 and Fusion 3. with 3Gb of RAM and 1 processor core dedicated to Fusion, Windows opens in less than a minute and I routinely leave it and Fusion running in Unity mode so that I can use Windows apps from the OS X desktop. I didn't even try to do that when my MBP had only 2Gb of RAM. Even now, Windows apps are not quite as fast as OS X apps but that was to be expected of any program running in a virtual environment. Nevertheless, the speed of Windows and its apps has been satisfactory.

My hard drive is the 160Gb Hitachi that was original equipment on my 2007 Macbook Pro.

Currently, on the iMac, it's running 3.0.2. XP is with 3GB w/ 1 Processor dedicated. It's hard drive is a 1.5TB Western Digital Green, 5400-7200RPM Variable. The MacBook Pro is running 3.1 Beta. Windows 7 is with 768MB w/ 1 Processor dedicated. The hard drive is 1.75TB Western Digital Blue (1TB + 750GB RAID-0), 5200RPM.

The MBP is also a Core Duo, the first generation 17" MBP while the iMac is a Core 2 Duo.

I use Outlook and Word 2007 in Unity a lot as I find Office 2008 to be extremely limited.

 

[gwsat]

It's in my sig. It's a 2.4GHz Core 2 Duo T7700 Processor.



Currently, on the iMac, it's running 3.0.2. XP is with 3GB w/ 1 Processor dedicated. It's hard drive is a 1.5TB Western Digital Green, 5400-7200RPM Variable. The MacBook Pro is running 3.1 Beta. Windows 7 is with 768MB w/ 1 Processor dedicated. The hard drive is 1.75TB Western Digital Blue (1TB + 750GB RAID-0), 5200RPM.

The MBP is also a Core Duo, the first generation 17" MBP while the iMac is a Core 2 Duo.

I use Outlook and Word 2007 in Unity a lot as I find Office 2008 to be extremely limited.

I see that you are running Fusion 3.1 beta, too. I guess that just goes to show that we are both gluttons for punishment. More seriously, I haven't had any significant trouble with Fusion 3.1 beta, although, as you no doubt are aware, others have. Your 2.4Ghz Core 2 Duo processor is apparently identical to the one in my MBP. I am now running Windows 7 instead of XP, though. Nevertheless, I can't account for the sluggishness you perceive when you use 3Gb of RAM. After I upped the memory available to Fusion to 3Gb from only 1Gb, both the speed of Windows apps and the stability of the virtual machine improved dramatically, although I was running everything in Unity mode. Windows apps aren't as fast as OS X apps but they are fast enough. I have been very pleased.

Most of the time I run only Quicken 2010 for Windows and The Oxford English Dictionary Version 2 on CD ROM. The OED is creaky but it still works -- sort of. I also run Wordperfect 12 occasionally because I still have a bunch of Wordperfect documents that I consult once in awhile. Although I used Outlook and Outlook Exchange for years, I have recently shifted to Apple Mail because I no longer have convenient access to an Outlook Exchange server.

 

[dyn]

I am convinced that the speed and stability of Windows apps running in Fusion on a Macbook Pro is mostly a function of how much memory you have. By current standards, my Macbook Pro's 2.4Ghz Core 2 Duo CPU is hardly the most capable kid on the block. Nevertheless, with 6Gb of RAM to work with, I could hardly ask for more when running Windows apps.

VMware uses technology that also heavily depends on the I/O speed. It simply reduces its memory footprint by swapping. It's swapping that benefits from more I/O speed. Memory is an important thing with virtualisation but not as much as I/O speed is. Using an ssd causes the machine to be a lot faster. In my case I could only run 2 vm's at a time with the hdd but 4 and even 6 with the ssd. We're talking Win XP, Win 7, Ubuntu 9.04 and Ubuntu 9.10. In both cases I had 4 GB of memory: 1 GB for Win 7, 512 for XP, Ubuntu 9.04, Ubuntu 9.10 and everything else for OS X. The cpu has hardly anything to do with vm's. My 2.0 GHz C2D in my Mac mini early 2009 couldn't go beyond 115% cpu usage, neither did the 2.4 GHz C2D in my MBP early 2008. The only bottleneck I now have is the 4 GB of memory. If I could up that to 8 I can assign more memory to my vm's or run even more vm's at the same time. The only thing is that the MBP is 2 years old, only works properly with 6 GB and the mini needs the very expensive 4 GB DDR3 SODIMM modules. I don't know if upgrading those machines can be justified (I can run 4 vm's at the same time by using both machines anyway).

 

[ayeying]

VMware uses technology that also heavily depends on the I/O speed. It simply reduces its memory footprint by swapping. It's swapping that benefits from more I/O speed. Memory is an important thing with virtualisation but not as much as I/O speed is. Using an ssd causes the machine to be a lot faster. In my case I could only run 2 vm's at a time with the hdd but 4 and even 6 with the ssd. We're talking Win XP, Win 7, Ubuntu 9.04 and Ubuntu 9.10. In both cases I had 4 GB of memory: 1 GB for Win 7, 512 for XP, Ubuntu 9.04, Ubuntu 9.10 and everything else for OS X. The cpu has hardly anything to do with vm's. My 2.0 GHz C2D in my Mac mini early 2009 couldn't go beyond 115% cpu usage, neither did the 2.4 GHz C2D in my MBP early 2008. The only bottleneck I now have is the 4 GB of memory. If I could up that to 8 I can assign more memory to my vm's or run even more vm's at the same time. The only thing is that the MBP is 2 years old, only works properly with 6 GB and the mini needs the very expensive 4 GB DDR3 SODIMM modules. I don't know if upgrading those machines can be justified (I can run 4 vm's at the same time by using both machines anyway).

If you add those up, you're left with 1.5GB of ram for OSX, which is plenty. On a more limited machine like my MacBook Pro, I'm virtually left with about 512-768MB of ram usually.

I do believe the I/O plays a huge role but I highly doubt that just switching to an SSD would help that much especially when ram would be more important when it comes to running multiple VMs.

 

[dyn]

Memory plays a huge part indeed but when it comes down to VMware products it doesn't play the biggest part. Slight difference in that. When you're running a vm it has some I/O but not much. If you are doing things like updating the vm, suspending, resuming it, a virus/malware scan or simply use some of the VMware Tools features this will create a lot of I/O. Don't forget that the entire hdd of a vm is also virtualised and causes quite some overhead and I/O.

There is also another part which is specifically something with VMware products. They tend to do quite some swapping which means a lot of I/O when things are getting written from memory to disk or vice versa (OS X also seems to do some swapping btw).

All in all this means that when you have a lot of vm's running the amount of I/O rises beyond the limits of ordinary hdd's causing a lot of slow downs or even bringing the entire system to a halt. The ssd's are very good at access times and random reads and writes making them very suited for things like virtualisation applications like VMware. This in turn causes a very noticeable speed increase allowing you to run more vm's and/or enjoy a more responsive system. It is the main reason why a lot virtualisation guys want the I/O virtualisation support the Core ix brings.

Below are two interesting articles that explain a bit about the importance of I/O:
- http://lonesysadmin.net/2006/05/20/vmware-io-problems/
- http://www.virtualization.info/2005/11/how-to-improve-disk-io-performances.html

Keep in mind that VMware uses about 90% of the same codebase in all of their products. So the explanation about Workstation will also apply to Player, Fusion, etc. Most other virtualisation products do somewhat same things to decrease their memory footprint and such.

 

[ayeying]

Memory plays a huge part indeed but when it comes down to VMware products it doesn't play the biggest part. Slight difference in that. When you're running a vm it has some I/O but not much. If you are doing things like updating the vm, suspending, resuming it, a virus/malware scan or simply use some of the VMware Tools features this will create a lot of I/O. Don't forget that the entire hdd of a vm is also virtualised and causes quite some overhead and I/O.

There is also another part which is specifically something with VMware products. They tend to do quite some swapping which means a lot of I/O when things are getting written from memory to disk or vice versa (OS X also seems to do some swapping btw).

All in all this means that when you have a lot of vm's running the amount of I/O rises beyond the limits of ordinary hdd's causing a lot of slow downs or even bringing the entire system to a halt. The ssd's are very good at access times and random reads and writes making them very suited for things like virtualisation applications like VMware. This in turn causes a very noticeable speed increase allowing you to run more vm's and/or enjoy a more responsive system. It is the main reason why a lot virtualisation guys want the I/O virtualisation support the Core ix brings.

Below are two interesting articles that explain a bit about the importance of I/O:
- http://lonesysadmin.net/2006/05/20/vmware-io-problems/
- http://www.virtualization.info/2005/11/how-to-improve-disk-io-performances.html

Keep in mind that VMware uses about 90% of the same codebase in all of their products. So the explanation about Workstation will also apply to Player, Fusion, etc. Most other virtualisation products do somewhat same things to decrease their memory footprint and such.

That's interesting. I never looked deeply into VMware's product's coding, but that's interesting how I/O plays a bigger role then ram.

I always assumed that ram was the real bottleneck since Workstation and Fusion seems to share most of the same coding, running VMs on Workstation is a bit more smooth (at least for me) then Fusion in comparison. I'm not saying it's horrible on either product but in my observations, workstation seems a bit better. Also, I've always found Workstation to be more a complete package and Fusion seems lacking in comparison to Workstation.

 

[dyn]

I thought it was all about memory as well until I actually tried the ssd. The amount of memory is useful so you can assign more memory to one or more vm's (which I like to do as Ubuntu 10.04 seems to need 1 GB of memory instead of the 512MB I gave it). I think the problem lies with the ordinary hdd, they are simply too slow for some stuff when compared to the ssd. But this doesn't mean memory isn't important or does not play a big role because it does.

I was trying to find the tech talk some VMware guy held a couple of years ago about VMware Fusion. It was quite interesting as he talked about the internals of Fusion, the VMware products and virtualisation. I managed to find it today so here it is: Fusion Geek Speak: Shawn Morel Talks About Virtualization at C4 Mac Dev Conference.

 

[gordonyz]

Guys

I want to know how the current P7550 (without vt-x) compared to P8700 on virtualization.

 

[ayeying]

I thought it was all about memory as well until I actually tried the ssd. The amount of memory is useful so you can assign more memory to one or more vm's (which I like to do as Ubuntu 10.04 seems to need 1 GB of memory instead of the 512MB I gave it). I think the problem lies with the ordinary hdd, they are simply too slow for some stuff when compared to the ssd. But this doesn't mean memory isn't important or does not play a big role because it does.

I was trying to find the tech talk some VMware guy held a couple of years ago about VMware Fusion. It was quite interesting as he talked about the internals of Fusion, the VMware products and virtualisation. I managed to find it today so here it is: Fusion Geek Speak: Shawn Morel Talks About Virtualization at C4 Mac Dev Conference.

Last time I used an SSD for virtualization, it caused problems with the Host OS. The Host kept stalling because the SSD couldn't keep up with the concurrent read and writes. For me, a HDD solves that problem. I'm not comparing read/write speed of SSD vs HDD here. I'm just saying, at least for me, SSD caused more issues then solving it.

 

[dyn]

That's a known problem with ssd's using a certain Jmicron controller (a lot of cheap ssd's use this controller). Ssd's with a controller from Samsung, Intel, Indilinx, Sandforce, etc. do not have these problems. There is also a known problem with the mid 2009 MacBook Pro with the system stalling for a second quite frequently but I think some software update fixed it (the problem arose with ssd's and hdd's, no matter what type or brand). In both cases the problems are hardware related and do not have anything to do with virtualisation or the host OS. Bit of a problem with ssd's: you need to do some research to get a proper ssd.

 

[ayeying]

That's a known problem with ssd's using a certain Jmicron controller (a lot of cheap ssd's use this controller). Ssd's with a controller from Samsung, Intel, Indilinx, Sandforce, etc. do not have these problems. There is also a known problem with the mid 2009 MacBook Pro with the system stalling for a second quite frequently but I think some software update fixed it (the problem arose with ssd's and hdd's, no matter what type or brand). In both cases the problems are hardware related and do not have anything to do with virtualisation or the host OS. Bit of a problem with ssd's: you need to do some research to get a proper ssd.

Yeah. that wasn't a choice on my end. The SSD I used was the one that came with the MacBook Air. Amazingly, there wasn't a pause of the host OS when it was booted into Windows, running VMware Workstation

 

[Vanarak19]

Keep VMs on a seperate fw800 HD? I used to do this. It worked out quite well. Internal HD was never limited, and I was only limited to do my VM work from my desk. Which in all honesty was ok, because it required a power cord.

If I ever went mobile, just copy the VM over, and leave.

 

[dusk007]

Here is something from anandtech to the topic.
And here an article about server CPUs that also cover Virtualization benchmarks.

Especially in VM Mark the Nehalem architecture destroyes the Penryns in performance. Apus says it is only about 35% difference, which is still huge.

In general with virtualization it is easy to run into some bottle neck like not enough RAM or too many IOs(less likely on usual load). But in terms of translating IOs, switching speed the Nehalem is much much faster than the old Penryn. More than 100% faster in VM Marks should tell you enough about how much has changed between those to architectures in this respect.