So I have a shiney new Sun Blade(X6270) dual socket Nehalem box with 24GB of RAM.
I loaded Solaris 10 update 7 + latest recommended. This puppy kicks some pretty serious butt! Running the patch cluster across it, it just flew by installing all the patches. I couldn't wait to get Vbox 3.0.8 up on it.
I did. And wow, what a horrid disappointment! It's so damn slow, feels like a 286 running Windows 3.1! It takes about 5 times longer to boot up than it does on my dual core Intel Laptop!
Something has to be wrong here...Any thoughts?
The laptop is running OpenSolaris 2009.06. The Nehalem is Sol 10/U7 + Patches
The laptop is running Vbox from local disk, the Nehalem from SAN disk
Both systems have VT-X enabled. I also tried nested paging on the Nelahem.
I also tried an AMD blade(X6220) and it was much faster than the Nehalem.
I then tried another Xeon blade, a X6250 with some dual core Xeons, and performance there is just as bad as the Nehalem.
So, why are the Xeons running Vbox so slow, when the AMD Operton and the Dual core Intel laptop are running just fine?
VBox and Nehalem, awful performance!
Re: VBox and Nehalem, awful performance!
I have the same experience. Horrid performance. My laptop is not quite as empowered as yours, however.
The only way I can get it to run quickly is to turn off I/O APIC.
That presents a problem, because it means you have to downrev your HAL from APIC to PIC. It's a problem, because I don't think you can easily go back to APIC should the performance thing get fixed.
The other difficulty with turning off I/O APIC, I think, is that I needed to downrev my kernel to ntoskrnl.exe, which is the uniprocessor kernel. I can't take advantage of the VT-x stuff, if there is an advantage.
It's disappointing to hear your spankin' new laptop is having the same performance experience as mine on the VT-x/APIC VMs. I was holding out hope that my older dual core laptop was just not up to the task and that the newer hardware would be better.
The only way I can get it to run quickly is to turn off I/O APIC.
That presents a problem, because it means you have to downrev your HAL from APIC to PIC. It's a problem, because I don't think you can easily go back to APIC should the performance thing get fixed.
The other difficulty with turning off I/O APIC, I think, is that I needed to downrev my kernel to ntoskrnl.exe, which is the uniprocessor kernel. I can't take advantage of the VT-x stuff, if there is an advantage.
It's disappointing to hear your spankin' new laptop is having the same performance experience as mine on the VT-x/APIC VMs. I was holding out hope that my older dual core laptop was just not up to the task and that the newer hardware would be better.
-
- Volunteer
- Posts: 3342
- Joined: 10. May 2007, 16:59
- Location: Israel
Re: VBox and Nehalem, awful performance!
1. Is your laptop running Solaris (host OS) ?
2. Try turn off nested paging
3. Try turn off hardware virtualization (VT-x)
2. Try turn off nested paging
3. Try turn off hardware virtualization (VT-x)
-
- Volunteer
- Posts: 1064
- Joined: 10. May 2007, 10:27
- Primary OS: MS Windows Vista
- VBox Version: PUEL
- Guest OSses: Windows, Linux, Solaris
Re: VBox and Nehalem, awful performance!
Well, my Core i7 3.2 Ghz system flies here with all guests. Nested paging gives a considerable performance boost as well. Include the VBox.log file of the 'horribly slow' session and I can have a quick look.
I'm still unsure which guest you're actually trying though. Solaris as well?
As for the IO-APIC overhead of the 2nd poster: your CPU doesn't support the APIC performance optimization and suffers badly with 32 bits Windows NT/XP/2003 guests. No such issues with Vista & Windows 7 or 64 bits guests. The next VirtualBox version will fix the IO-APIC overhead for Intel CPUs that support 64 bits mode. (leaves out some Core Solo/Duo mobile CPUs)
I'm still unsure which guest you're actually trying though. Solaris as well?
As for the IO-APIC overhead of the 2nd poster: your CPU doesn't support the APIC performance optimization and suffers badly with 32 bits Windows NT/XP/2003 guests. No such issues with Vista & Windows 7 or 64 bits guests. The next VirtualBox version will fix the IO-APIC overhead for Intel CPUs that support 64 bits mode. (leaves out some Core Solo/Duo mobile CPUs)
-
- Volunteer
- Posts: 1003
- Joined: 5. Sep 2008, 14:40
- Primary OS: Debian other
- VBox Version: PUEL
- Guest OSses: Solaris, Linux, Windows, OS/2, MacOSX, FreeBSD
- Contact:
Re: VBox and Nehalem, awful performance!
is the io-apic cpu support what is called "VT-d"? as i was thinking of buying a core-i5 750 but it doesn't have VT-d support unlike the core-i7 (it has VT-x obviously) and i believe it has nested paging, although i can't find a reliable list of which processors support NPT.sandervl wrote:Well, my Core i7 3.2 Ghz system flies here with all guests. Nested paging gives a considerable performance boost as well. Include the VBox.log file of the 'horribly slow' session and I can have a quick look.
I'm still unsure which guest you're actually trying though. Solaris as well?
As for the IO-APIC overhead of the 2nd poster: your CPU doesn't support the APIC performance optimization and suffers badly with 32 bits Windows NT/XP/2003 guests. No such issues with Vista & Windows 7 or 64 bits guests. The next VirtualBox version will fix the IO-APIC overhead for Intel CPUs that support 64 bits mode. (leaves out some Core Solo/Duo mobile CPUs)
i seem to be able to use io-apic on my core2quad q6600....
-
- Volunteer
- Posts: 1064
- Joined: 10. May 2007, 10:27
- Primary OS: MS Windows Vista
- VBox Version: PUEL
- Guest OSses: Windows, Linux, Solaris
Re: VBox and Nehalem, awful performance!
No, VT-d is the hardware extension to directly use (pci) devices in the VM. The APIC optimization is the VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC capability (can be found in VBox.log).sej7278 wrote: is the io-apic cpu support what is called "VT-d"? as i was thinking of buying a core-i5 750 but it doesn't have VT-d support unlike the core-i7 (it has VT-x obviously) and i believe it has nested paging, although i can't find a reliable list of which processors support NPT.
i seem to be able to use io-apic on my core2quad q6600....
-
- Volunteer
- Posts: 1003
- Joined: 5. Sep 2008, 14:40
- Primary OS: Debian other
- VBox Version: PUEL
- Guest OSses: Solaris, Linux, Windows, OS/2, MacOSX, FreeBSD
- Contact:
Re: VBox and Nehalem, awful performance!
ah that's very useful thanks - so i might as well stick with an i5, assuming its has nested paging (i think all nehalems have that as part of VT-x?) as i'm not interested in pci-passthru.sandervl wrote:No, VT-d is the hardware extension to directly use (pci) devices in the VM. The APIC optimization is the VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC capability (can be found in VBox.log).
-
- Volunteer
- Posts: 1064
- Joined: 10. May 2007, 10:27
- Primary OS: MS Windows Vista
- VBox Version: PUEL
- Guest OSses: Windows, Linux, Solaris
Re: VBox and Nehalem, awful performance!
That's my guess as well, but I really recommend checking this very carefully as Intel is notorious for shipping CPUs with different capabilities.sej7278 wrote:ah that's very useful thanks - so i might as well stick with an i5, assuming its has nested paging (i think all nehalems have that as part of VT-x?) as i'm not interested in pci-passthru.sandervl wrote:No, VT-d is the hardware extension to directly use (pci) devices in the VM. The APIC optimization is the VMX_VMCS_CTRL_PROC_EXEC2_VIRT_APIC capability (can be found in VBox.log).
Re: VBox and Nehalem, awful performance!
Ok, so my laptop has a Core 2 Duo T5600 1.83 GHz Merom chip, and the host OS is 64-bit OpenSolaris 2009.06. The guest is 32-bit XP. You are saying that the APIC thing is a "feature" of my older chip when going from 64 bit host to 32 bit guest. You are also saying this "feature" will be compensated for in the next version of VirtualBox.sandervl wrote:As for the IO-APIC overhead of the 2nd poster: your CPU doesn't support the APIC performance optimization and suffers badly with 32 bits Windows NT/XP/2003 guests. No such issues with Vista & Windows 7 or 64 bits guests. The next VirtualBox version will fix the IO-APIC overhead for Intel CPUs that support 64 bits mode. (leaves out some Core Solo/Duo mobile CPUs)
Do I have it right? Or should I just plan on switching to a Vista/Win7 guest anyway Real Soon Now?
And was this "feature" written up in a release note somewhere that I missed?
-
- Volunteer
- Posts: 1064
- Joined: 10. May 2007, 10:27
- Primary OS: MS Windows Vista
- VBox Version: PUEL
- Guest OSses: Windows, Linux, Solaris
Re: VBox and Nehalem, awful performance!
No, not quite. Your CPU does not have this hardware feature so VirtualBox has to work around this. This workaround requires 64 bits support, which your CPU offers.pel wrote:Ok, so my laptop has a Core 2 Duo T5600 1.83 GHz Merom chip, and the host OS is 64-bit OpenSolaris 2009.06. The guest is 32-bit XP. You are saying that the APIC thing is a "feature" of my older chip when going from 64 bit host to 32 bit guest. You are also saying this "feature" will be compensated for in the next version of VirtualBox.sandervl wrote:As for the IO-APIC overhead of the 2nd poster: your CPU doesn't support the APIC performance optimization and suffers badly with 32 bits Windows NT/XP/2003 guests. No such issues with Vista & Windows 7 or 64 bits guests. The next VirtualBox version will fix the IO-APIC overhead for Intel CPUs that support 64 bits mode. (leaves out some Core Solo/Duo mobile CPUs)
Do I have it right? Or should I just plan on switching to a Vista/Win7 guest anyway Real Soon Now?
And was this "feature" written up in a release note somewhere that I missed?
This limitation is documented in the manual.
The next VirtualBox version will include it.