Hi, I measured network performance with netperf on PV-on-HVM: Guest Kernel is taken from git:// xenbits.xen.org/people/sstabellini/linux-pvhvm.git. Xen hypervisor is taken from http://koji.fedoraproject.org/koji/taskinfo?taskID=2478501. Dom0 Kernel is taken from git:// git.kernel.org/pub/scm/linux/kernel/git/jermy/xen.git and switched to stable branch xen/stable-2.6.32.x). The Kernel configuration is taken from http://pasik.reaktio.net/xen/pv_ops-dom0-debug/config-2.6.32.21-pvops-dom0-xen-stable-x86_64. The tutorial mentioned that the Kernel configuration has debug enabled and there will be a performance hit but I first wanted to get the setup UP and get some performance numbers(I will get the performance numbers by disabling the debug parameters in Kernel config and I will let you know those performance numbers). With the above combination, I am able to get the PV-on-HVM guest up and running. My machine has the following harware configuration. (1) Chipset - Intel 82Q35 (2) Processor - Intel Core2 Duo CPU E8400 @ 3.00GHz Performance setup: =============The dom0''s Ethernet interface is connected to LAN. The netserver is running on another host that is connected to LAN. The netperf client is running on the PV-on-HVM guest. So, we are measuring performance from PV-on-HVM guest to a host on LAN. Using netperf, the performance numbers in Mbps are 723, 719, 703, 735, 709, 733 for 6 trails. Are there any published performance numbers with PV-on-HVM guest? I want to compare my numbers with the published ones. Thanks & Regards, Jagadish _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Sat, 25 Sep 2010, Jagadish Nadimpalli wrote:> Hi, > > I measured network performance with netperf on PV-on-HVM: > Guest Kernel is taken from git://xenbits.xen.org/people/sstabellini/linux-pvhvm.git.The most up to date branch is 2.6.35-rc5-pvhvm-v7, please use it for your tests. Also make sure that the PV interfaces are actually running, in particular use the right name for the root device (''/dev/xvda'', if you are using ''hda'' in your VM config file).> Xen hypervisor is taken from http://koji.fedoraproject.org/koji/taskinfo?taskID=2478501.I couldn''t find the download link for xen-4.0.1-1.fc13.1.src.rpm but I found the download link for xen-4.0.1-1.fc14.1.src.rpm and it contains all the bits needed.> Dom0 Kernel is taken from git://git.kernel.org/pub/scm/linux/kernel/git/jermy/xen.git and switched to stable branch > xen/stable-2.6.32.x). > The Kernel configuration is taken from > http://pasik.reaktio.net/xen/pv_ops-dom0-debug/config-2.6.32.21-pvops-dom0-xen-stable-x86_64. > > The tutorial mentioned that the Kernel configuration has debug enabled and there will be a performance hit but I first > wanted to get the setup UP and get some performance numbers(I will get the performance numbers by disabling the debug > parameters in Kernel config and I will let you know those performance numbers). With the above combination, I am able to > get the PV-on-HVM guest up and running. > > My machine has the following harware configuration. > (1) Chipset - Intel 82Q35 > (2) Processor - Intel Core2 Duo CPU E8400 @ 3.00GHz >Performance results might vary significantly depending on whether the cpu supports EPT or not. Could you please cat /proc/cpuinfo? If this cpu doesn''t support EPT, it might be worth changing test hardware :)> Performance setup: > =============> The dom0''s Ethernet interface is connected to LAN. The netserver is running on another host that is connected to LAN. The > netperf client is running on the PV-on-HVM guest. So, we are measuring performance from PV-on-HVM guest to a host on LAN. > Using netperf, the performance numbers in Mbps are 723, 719, 703, 735, 709, 733 for 6 trails. > > Are there any published performance numbers with PV-on-HVM guest? I want to compare my numbers with the published ones. >I ran few kernbench a while back, these are the test results: http://lists.xensource.com/archives/html/xen-devel/2010-05/msg01176.html _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jagadish Nadimpalli
2010-Sep-27 11:27 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
Hi Stefano, Thanks for your response. Please find my responses inline. On Mon, Sep 27, 2010 at 3:52 PM, Stefano Stabellini < stefano.stabellini@eu.citrix.com> wrote:> On Sat, 25 Sep 2010, Jagadish Nadimpalli wrote: > > Hi, > > > > I measured network performance with netperf on PV-on-HVM: > > Guest Kernel is taken from git:// > xenbits.xen.org/people/sstabellini/linux-pvhvm.git. > > The most up to date branch is 2.6.35-rc5-pvhvm-v7, please use it for > your tests. >I will use the latest branch. Also make sure that the PV interfaces are actually running, in> particular use the right name for the root device (''/dev/xvda'', > if you are using ''hda'' in your VM config file). >Could you please let me know where(config file) can I configure/verify the name for root device. Since I am testing network performance, I verified which network interface is used in the PV-on-HVM VM. I found that there is no Ethernet device displayed in the lspci output of VM. Here is the lspci output on PV-on-HVM VM. The Xen platform device is displayed in lspci. This means that the platform device is used as PV driver in the PV-on-HVM guest. Please correct me if I am wrong. ==========================================================00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02) 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II] 00:01.2 USB Controller: Intel Corporation 82371SB PIIX3 USB [Natoma/Triton II] (rev 01) 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 01) 00:02.0 VGA compatible controller: Cirrus Logic GD 5446 00:03.0 Class ff80: XenSource, Inc. Xen Platform Device (rev 01) 00:05.0 Multimedia audio controller: Ensoniq ES1370 [AudioPCI] ==========================================================> > > Xen hypervisor is taken from > http://koji.fedoraproject.org/koji/taskinfo?taskID=2478501. > > I couldn''t find the download link for xen-4.0.1-1.fc13.1.src.rpm but I > found the download link for xen-4.0.1-1.fc14.1.src.rpm and it contains > all the bits needed. > > > > Dom0 Kernel is taken from git:// > git.kernel.org/pub/scm/linux/kernel/git/jermy/xen.git and switched to > stable branch > > xen/stable-2.6.32.x). > > The Kernel configuration is taken from > > > http://pasik.reaktio.net/xen/pv_ops-dom0-debug/config-2.6.32.21-pvops-dom0-xen-stable-x86_64 > . > > > > The tutorial mentioned that the Kernel configuration has debug enabled > and there will be a performance hit but I first > > wanted to get the setup UP and get some performance numbers(I will get > the performance numbers by disabling the debug > > parameters in Kernel config and I will let you know those performance > numbers). With the above combination, I am able to > > get the PV-on-HVM guest up and running. > > > > My machine has the following harware configuration. > > (1) Chipset - Intel 82Q35 > > (2) Processor - Intel Core2 Duo CPU E8400 @ 3.00GHz > > > > Performance results might vary significantly depending on whether the > cpu supports EPT or not. Could you please cat /proc/cpuinfo? > If this cpu doesn''t support EPT, it might be worth changing test > hardware :) >Here is the cat /proc/cpuinfo output: =================================================processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 23 model name : Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz stepping : 10 cpu MHz : 2992.566 cache size : 6144 KB physical id : 0 siblings : 2 core id : 0 cpu cores : 1 apicid : 0 initial apicid : 0 fpu : yes fpu_exception : yes cpuid level : 2 wp : yes flags : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat clflush acpi mmx fxsr sse sse2 ss ht syscall nx lm constant_tsc rep_good pni est ssse3 cx16 sse4_1 hypervisor lahf_lm bogomips : 5985.13 clflush size : 64 cache_alignment : 64 address sizes : 36 bits physical, 48 bits virtual power management: processor : 1 vendor_id : GenuineIntel cpu family : 6 model : 23 model name : Intel(R) Core(TM)2 Duo CPU E8400 @ 3.00GHz stepping : 10 cpu MHz : 2992.566 cache size : 6144 KB physical id : 0 siblings : 2 core id : 0 cpu cores : 1 apicid : 1 initial apicid : 1 fpu : yes fpu_exception : yes cpuid level : 2 wp : yes flags : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat clflush acpi mmx fxsr sse sse2 ss ht syscall nx lm constant_tsc rep_good pni est ssse3 cx16 sse4_1 hypervisor lahf_lm bogomips : 5985.13 clflush size : 64 cache_alignment : 64 address sizes : 36 bits physical, 48 bits virtual power management: ============================================================> > > > Performance setup: > > =============> > The dom0''s Ethernet interface is connected to LAN. The netserver is > running on another host that is connected to LAN. The > > netperf client is running on the PV-on-HVM guest. So, we are measuring > performance from PV-on-HVM guest to a host on LAN. > > Using netperf, the performance numbers in Mbps are 723, 719, 703, 735, > 709, 733 for 6 trails. > > > > Are there any published performance numbers with PV-on-HVM guest? I want > to compare my numbers with the published ones. > > > > I ran few kernbench a while back, these are the test results: > > http://lists.xensource.com/archives/html/xen-devel/2010-05/msg01176.html > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote:> Also make sure that the PV interfaces are actually running, in > particular use the right name for the root device (''/dev/xvda'', > if you are using ''hda'' in your VM config file). > > ?? > Could you please let me know where(config file) can I configure/verify the name for root device. > ??A regular hvm config file should work fine, a disk configuration like the following should work correctly: disk = [ ''file:/root/images/lenny.img,hda:disk,w'' ] You can manually specify the root device using root= on the kernel command line, for example root=/dev/xvda1. If everything worked as expected, you should have something like this in your dmesg: blkfront: xvda: barriers enabled ide0: BM-DMA at 0xc140-0xc147 ide1: BM-DMA at 0xc148-0xc14f Probing IDE interface ide0... xvda: xvda1 xvda2 < xvda5 > and you should have xvda under /dev but /dev/hda should be missing.> Since I am testing network performance, I verified which network interface is used in the PV-on-HVM VM. I found that there > is no Ethernet device displayed in the lspci output of VM. Here is the lspci output on PV-on-HVM VM. The Xen platform > device is displayed in lspci. This means that the platform device is used as PV driver in the PV-on-HVM guest. Please > correct me if I am wrong. > > ==========================================================> 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02) > 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II] > 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II] > 00:01.2 USB Controller: Intel Corporation 82371SB PIIX3 USB [Natoma/Triton II] (rev 01) > 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 01) > 00:02.0 VGA compatible controller: Cirrus Logic GD 5446 > 00:03.0 Class ff80: XenSource, Inc. Xen Platform Device (rev 01) > 00:05.0 Multimedia audio controller: Ensoniq ES1370 [AudioPCI] > ==========================================================> >this is a good sign: it means that the emulated pci network device has been unplugged because the kernel setup netfront instead.> > My machine has the following harware configuration. > > (1) Chipset - Intel 82Q35 > > (2) Processor - Intel Core2 Duo CPU E8400 @ 3.00GHz > > > > Performance results might vary significantly depending on whether the > cpu supports EPT or not. Could you please cat /proc/cpuinfo? > If this cpu doesn''t support EPT, it might be worth changing test > hardware :) > > ?? > Here is the cat /proc/cpuinfo output: > =================================================> processor?????? : 0 > vendor_id?????? : GenuineIntel > cpu family?????? : 6 > model?????? ?????? : 23 > model name?????? : Intel(R) Core(TM)2 Duo CPU???????? E8400?? @ 3.00GHz > stepping?????? : 10 > cpu MHz?????? ?????? : 2992.566 > cache size?????? : 6144 KB > physical id?????? : 0 > siblings?????? : 2 > core id?????? ?????? : 0 > cpu cores?????? : 1 > apicid?????? ?????? : 0 > initial apicid?????? : 0 > fpu?????? ?????? : yes > fpu_exception?????? : yes > cpuid level?????? : 2 > wp?????? ?????? : yes > flags?????? ?????? : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat clflush acpi mmx fxsr sse sse2 ss ht syscall nx lm > constant_tsc rep_good pni est ssse3 cx16 sse4_1 hypervisor lahf_lm > bogomips?????? : 5985.13 > clflush size?????? : 64 > cache_alignment?????? : 64 > address sizes?????? : 36 bits physical, 48 bits virtual > power management: > > processor?????? : 1 > vendor_id?????? : GenuineIntel > cpu family?????? : 6 > model?????? ?????? : 23 > model name?????? : Intel(R) Core(TM)2 Duo CPU???????? E8400?? @ 3.00GHz > stepping?????? : 10 > cpu MHz?????? ?????? : 2992.566 > cache size?????? : 6144 KB > physical id?????? : 0 > siblings?????? : 2 > core id?????? ?????? : 0 > cpu cores?????? : 1 > apicid?????? ?????? : 1 > initial apicid?????? : 1 > fpu?????? ?????? : yes > fpu_exception?????? : yes > cpuid level?????? : 2 > wp?????? ?????? : yes > flags?????? ?????? : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov pat clflush acpi mmx fxsr sse sse2 ss ht syscall nx lm > constant_tsc rep_good pni est ssse3 cx16 sse4_1 hypervisor lahf_lm > bogomips?????? : 5985.13 > clflush size?????? : 64 > cache_alignment?????? : 64 > address sizes?????? : 36 bits physical, 48 bits virtual > power management: > ============================================================> ??Sorry I have just realized that if you executed cat /proc/cpuinfo from dom0 under xen the hap flag would be missing anyway. Did xen printed out "Extended Page Tables (EPT)" at boot? You can also check the xen output executing xm dmesg from dom0 after host boot. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jagadish Nadimpalli
2010-Sep-27 12:42 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
Hi Stefano, Please find my responses inline. On Mon, Sep 27, 2010 at 5:19 PM, Stefano Stabellini < stefano.stabellini@eu.citrix.com> wrote:> On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote: > > Also make sure that the PV interfaces are actually running, in > > particular use the right name for the root device (''/dev/xvda'', > > if you are using ''hda'' in your VM config file). > > > > ?? > > Could you please let me know where(config file) can I configure/verify > the name for root device. > > ?? > > A regular hvm config file should work fine, a disk configuration like > the following should work correctly: > > disk = [ ''file:/root/images/lenny.img,hda:disk,w'' ] > > You can manually specify the root device using root= on the kernel > command line, for example root=/dev/xvda1. > If everything worked as expected, you should have something like this in > your dmesg: > > blkfront: xvda: barriers enabled > ide0: BM-DMA at 0xc140-0xc147 > ide1: BM-DMA at 0xc148-0xc14f > Probing IDE interface ide0... > xvda: xvda1 xvda2 < xvda5 > > > and you should have xvda under /dev but /dev/hda should be missing. > > > Since I am testing network performance, I verified which network > interface is used in the PV-on-HVM VM. I found that there > > is no Ethernet device displayed in the lspci output of VM. Here is the > lspci output on PV-on-HVM VM. The Xen platform > > device is displayed in lspci. This means that the platform device is used > as PV driver in the PV-on-HVM guest. Please > > correct me if I am wrong. > > > > ==========================================================> > 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev > 02) > > 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton > II] > > 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton > II] > > 00:01.2 USB Controller: Intel Corporation 82371SB PIIX3 USB > [Natoma/Triton II] (rev 01) > > 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 01) > > 00:02.0 VGA compatible controller: Cirrus Logic GD 5446 > > 00:03.0 Class ff80: XenSource, Inc. Xen Platform Device (rev 01) > > 00:05.0 Multimedia audio controller: Ensoniq ES1370 [AudioPCI] > > ==========================================================> > > > > > this is a good sign: it means that the emulated pci network device has > been unplugged because the kernel setup netfront instead. >Did anybody calculate benchmark network performance numbers for PV-on-HVM so that I can compare? The link provided by you has some numbers as follows. This don''t has the data rate that is transferred via a network interface. *Average Optimal load -j 4 Run (std deviation): Elapsed Time 215.307 (10.1294) User Time 632.503 (6.4785) System Time 115.497 (4.53905) Percent CPU 347.333 (15.885) Context Switches 43319.7 (2088.39) Sleeps 48950 (3140.18) *> > > > My machine has the following harware configuration. > > > (1) Chipset - Intel 82Q35 > > > (2) Processor - Intel Core2 Duo CPU E8400 @ 3.00GHz > > > > > > > Performance results might vary significantly depending on whether the > > cpu supports EPT or not. Could you please cat /proc/cpuinfo? > > If this cpu doesn''t support EPT, it might be worth changing test > > hardware :) > > > > ?? > > Here is the cat /proc/cpuinfo output: > > =================================================> > processor?????? : 0 > > vendor_id?????? : GenuineIntel > > cpu family?????? : 6 > > model?????? ?????? : 23 > > model name?????? : Intel(R) Core(TM)2 Duo CPU???????? E8400?? @ 3.00GHz > > stepping?????? : 10 > > cpu MHz?????? ?????? : 2992.566 > > cache size?????? : 6144 KB > > physical id?????? : 0 > > siblings?????? : 2 > > core id?????? ?????? : 0 > > cpu cores?????? : 1 > > apicid?????? ?????? : 0 > > initial apicid?????? : 0 > > fpu?????? ?????? : yes > > fpu_exception?????? : yes > > cpuid level?????? : 2 > > wp?????? ?????? : yes > > flags?????? ?????? : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov > pat clflush acpi mmx fxsr sse sse2 ss ht syscall nx lm > > constant_tsc rep_good pni est ssse3 cx16 sse4_1 hypervisor lahf_lm > > bogomips?????? : 5985.13 > > clflush size?????? : 64 > > cache_alignment?????? : 64 > > address sizes?????? : 36 bits physical, 48 bits virtual > > power management: > > > > processor?????? : 1 > > vendor_id?????? : GenuineIntel > > cpu family?????? : 6 > > model?????? ?????? : 23 > > model name?????? : Intel(R) Core(TM)2 Duo CPU???????? E8400?? @ 3.00GHz > > stepping?????? : 10 > > cpu MHz?????? ?????? : 2992.566 > > cache size?????? : 6144 KB > > physical id?????? : 0 > > siblings?????? : 2 > > core id?????? ?????? : 0 > > cpu cores?????? : 1 > > apicid?????? ?????? : 1 > > initial apicid?????? : 1 > > fpu?????? ?????? : yes > > fpu_exception?????? : yes > > cpuid level?????? : 2 > > wp?????? ?????? : yes > > flags?????? ?????? : fpu de tsc msr pae mce cx8 apic sep mtrr mca cmov > pat clflush acpi mmx fxsr sse sse2 ss ht syscall nx lm > > constant_tsc rep_good pni est ssse3 cx16 sse4_1 hypervisor lahf_lm > > bogomips?????? : 5985.13 > > clflush size?????? : 64 > > cache_alignment?????? : 64 > > address sizes?????? : 36 bits physical, 48 bits virtual > > power management: > > ============================================================> > ?? > > > Sorry I have just realized that if you executed cat /proc/cpuinfo from > dom0 under xen the hap flag would be missing anyway. > Did xen printed out "Extended Page Tables (EPT)" at boot? > You can also check the xen output executing xm dmesg from dom0 after > host boot. >I verified the "xm dmesg" and there is no print of "Extended Page Tables (EPT)". Does this mean that the extended page table support is not there? Can I enable this through BIOS option? If it can''t be enabled using BIOS option, could you please let me know the Intel chipset series which has this support. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote:> Did anybody calculate benchmark network performance numbers for PV-on-HVM so that I can compare? The link provided by you > has some numbers as follows. This don''t has the data rate that is transferred via a network interface. > > Average Optimal load -j 4 Run (std deviation): > Elapsed Time 215.307 (10.1294) > User Time 632.503 (6.4785) > System Time 115.497 (4.53905) > Percent CPU 347.333 (15.885) > Context Switches 43319.7 (2088.39) > > Sleeps 48950 (3140.18) >I don''t think anyone did so far.> ?? > ?? > I verified the "xm dmesg" and there is no print of "Extended Page Tables (EPT)". Does this mean that the extended page > table support is not there? Can I enable this through BIOS option? If it can''t be enabled using BIOS option, could you > please let me know the Intel chipset series which has this support. > >I don''t think that EPT can be enable/disable via BIOS, probably your cpu doesn''t support it. The first Intel cpu series to support EPT is Nehalem: http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29 all the new desktops and servers sold today should have ept support, you can check on the intel website to be sure. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jagadish Nadimpalli
2010-Sep-27 15:48 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
Hi Stefano, Thanks a lot for your responses. Thanks & Regards, Jagadish On Mon, Sep 27, 2010 at 6:41 PM, Stefano Stabellini < stefano.stabellini@eu.citrix.com> wrote:> On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote: > > Did anybody calculate benchmark network performance numbers for PV-on-HVM > so that I can compare? The link provided by you > > has some numbers as follows. This don''t has the data rate that is > transferred via a network interface. > > > > Average Optimal load -j 4 Run (std deviation): > > Elapsed Time 215.307 (10.1294) > > User Time 632.503 (6.4785) > > System Time 115.497 (4.53905) > > Percent CPU 347.333 (15.885) > > Context Switches 43319.7 (2088.39) > > > > Sleeps 48950 (3140.18) > > > > I don''t think anyone did so far. > > > ?? > > ?? > > I verified the "xm dmesg" and there is no print of "Extended Page Tables > (EPT)". Does this mean that the extended page > > table support is not there? Can I enable this through BIOS option? If it > can''t be enabled using BIOS option, could you > > please let me know the Intel chipset series which has this support. > > > > > > I don''t think that EPT can be enable/disable via BIOS, probably your > cpu doesn''t support it. > The first Intel cpu series to support EPT is Nehalem: > > http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29 > > all the new desktops and servers sold today should have ept support, you > can check on the intel website to be sure. >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jagadish Nadimpalli
2010-Sep-28 10:28 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
Hi, With latest PV-on-HVM branch 2.6.35-rc5-pvhvm-v7, the network performance results(Using netperf -H <IP address of netserver> -l 100) range from 700Mbps to 730Mbps. Please note that this is similar to the old branch 2.6.34-pvhvm. One difference is that the latest branch is much more stable. The old branch code used to freeze a lot. Thanks & Regards, Jagadish On Mon, Sep 27, 2010 at 9:18 PM, Jagadish Nadimpalli < jagadish.nadimpalli@oneconvergence.com> wrote:> Hi Stefano, > > Thanks a lot for your responses. > > Thanks & Regards, > Jagadish > > > On Mon, Sep 27, 2010 at 6:41 PM, Stefano Stabellini < > stefano.stabellini@eu.citrix.com> wrote: > >> On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote: >> > Did anybody calculate benchmark network performance numbers for >> PV-on-HVM so that I can compare? The link provided by you >> > has some numbers as follows. This don''t has the data rate that is >> transferred via a network interface. >> > >> > Average Optimal load -j 4 Run (std deviation): >> > Elapsed Time 215.307 (10.1294) >> > User Time 632.503 (6.4785) >> > System Time 115.497 (4.53905) >> > Percent CPU 347.333 (15.885) >> > Context Switches 43319.7 (2088.39) >> > >> > Sleeps 48950 (3140.18) >> > >> >> I don''t think anyone did so far. >> >> > ?? >> > ?? >> > I verified the "xm dmesg" and there is no print of "Extended Page Tables >> (EPT)". Does this mean that the extended page >> > table support is not there? Can I enable this through BIOS option? If it >> can''t be enabled using BIOS option, could you >> > please let me know the Intel chipset series which has this support. >> > >> > >> >> I don''t think that EPT can be enable/disable via BIOS, probably your >> cpu doesn''t support it. >> The first Intel cpu series to support EPT is Nehalem: >> >> http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29 >> >> all the new desktops and servers sold today should have ept support, you >> can check on the intel website to be sure. >> > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Tue, Sep 28, 2010 at 03:58:50PM +0530, Jagadish Nadimpalli wrote:> Hi, > > With latest PV-on-HVM branch 2.6.35-rc5-pvhvm-v7, the network performance > results(Using netperf -H <IP address of netserver> -l 100) range from > 700Mbps to 730Mbps. Please note that this is similar to the old branch > 2.6.34-pvhvm. One difference is that the latest branch is much more > stable. The old branch code used to freeze a lot. >These benchmark results are still on a non-EPT CPU? -- Pasi> Thanks & Regards, > Jagadish > > On Mon, Sep 27, 2010 at 9:18 PM, Jagadish Nadimpalli > <[1]jagadish.nadimpalli@oneconvergence.com> wrote: > > Hi Stefano, > > Thanks a lot for your responses. > > Thanks & Regards, > Jagadish > > On Mon, Sep 27, 2010 at 6:41 PM, Stefano Stabellini > <[2]stefano.stabellini@eu.citrix.com> wrote: > > On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote: > > Did anybody calculate benchmark network performance numbers for > PV-on-HVM so that I can compare? The link provided by you > > has some numbers as follows. This don''t has the data rate that is > transferred via a network interface. > > > > Average Optimal load -j 4 Run (std deviation): > > Elapsed Time 215.307 (10.1294) > > User Time 632.503 (6.4785) > > System Time 115.497 (4.53905) > > Percent CPU 347.333 (15.885) > > Context Switches 43319.7 (2088.39) > > > > Sleeps 48950 (3140.18) > > > > I don''t think anyone did so far. > > > ?? > > ?? > > I verified the "xm dmesg" and there is no print of "Extended Page > Tables (EPT)". Does this mean that the extended page > > table support is not there? Can I enable this through BIOS option? > If it can''t be enabled using BIOS option, could you > > please let me know the Intel chipset series which has this support. > > > > > > I don''t think that EPT can be enable/disable via BIOS, probably your > cpu doesn''t support it. > The first Intel cpu series to support EPT is Nehalem: > > [3]http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29 > > all the new desktops and servers sold today should have ept support, > you > can check on the intel website to be sure. > > References > > Visible links > 1. mailto:jagadish.nadimpalli@oneconvergence.com > 2. mailto:stefano.stabellini@eu.citrix.com > 3. http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29> _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jagadish Nadimpalli
2010-Sep-28 11:28 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
On Tue, Sep 28, 2010 at 4:31 PM, Pasi Kärkkäinen <pasik@iki.fi> wrote:> On Tue, Sep 28, 2010 at 03:58:50PM +0530, Jagadish Nadimpalli wrote: > > Hi, > > > > With latest PV-on-HVM branch 2.6.35-rc5-pvhvm-v7, the network > performance > > results(Using netperf -H <IP address of netserver> -l 100) range from > > 700Mbps to 730Mbps. Please note that this is similar to the old branch > > 2.6.34-pvhvm. One difference is that the latest branch is much more > > stable. The old branch code used to freeze a lot. > > > > These benchmark results are still on a non-EPT CPU? >Yes. These results are still on a non-EPT CPU.> > -- Pasi > > > Thanks & Regards, > > Jagadish > > > > On Mon, Sep 27, 2010 at 9:18 PM, Jagadish Nadimpalli > > <[1]jagadish.nadimpalli@oneconvergence.com> wrote: > > > > Hi Stefano, > > > > Thanks a lot for your responses. > > > > Thanks & Regards, > > Jagadish > > > > On Mon, Sep 27, 2010 at 6:41 PM, Stefano Stabellini > > <[2]stefano.stabellini@eu.citrix.com> wrote: > > > > On Mon, 27 Sep 2010, Jagadish Nadimpalli wrote: > > > Did anybody calculate benchmark network performance numbers for > > PV-on-HVM so that I can compare? The link provided by you > > > has some numbers as follows. This don''t has the data rate that > is > > transferred via a network interface. > > > > > > Average Optimal load -j 4 Run (std deviation): > > > Elapsed Time 215.307 (10.1294) > > > User Time 632.503 (6.4785) > > > System Time 115.497 (4.53905) > > > Percent CPU 347.333 (15.885) > > > Context Switches 43319.7 (2088.39) > > > > > > Sleeps 48950 (3140.18) > > > > > > > I don''t think anyone did so far. > > > > > ?? > > > ?? > > > I verified the "xm dmesg" and there is no print of "Extended > Page > > Tables (EPT)". Does this mean that the extended page > > > table support is not there? Can I enable this through BIOS > option? > > If it can''t be enabled using BIOS option, could you > > > please let me know the Intel chipset series which has this > support. > > > > > > > > > > I don''t think that EPT can be enable/disable via BIOS, probably > your > > cpu doesn''t support it. > > The first Intel cpu series to support EPT is Nehalem: > > > > [3]http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29 > > > > all the new desktops and servers sold today should have ept > support, > > you > > can check on the intel website to be sure. > > > > References > > > > Visible links > > 1. mailto:jagadish.nadimpalli@oneconvergence.com > > 2. mailto:stefano.stabellini@eu.citrix.com > > 3. http://en.wikipedia.org/wiki/Nehalem_%28microarchitecture%29 > > > _______________________________________________ > > Xen-devel mailing list > > Xen-devel@lists.xensource.com > > http://lists.xensource.com/xen-devel > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Il 27/09/2010 12:22, Stefano Stabellini ha scritto:> On Sat, 25 Sep 2010, Jagadish Nadimpalli wrote: >> Hi, >> >> I measured network performance with netperf on PV-on-HVM: >> Guest Kernel is taken from git://xenbits.xen.org/people/sstabellini/linux-pvhvm.git. > > The most up to date branch is 2.6.35-rc5-pvhvm-v7, please use it for > your tests.Hi, do you plan to backport your changes on a widely adopted long time supported 2.6.32.x? if you backport the changes, more people could test and bench it on different hardwares Best regards, Christian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Tue, 28 Sep 2010, Christian Zoffoli wrote:> Il 27/09/2010 12:22, Stefano Stabellini ha scritto: > > On Sat, 25 Sep 2010, Jagadish Nadimpalli wrote: > >> Hi, > >> > >> I measured network performance with netperf on PV-on-HVM: > >> Guest Kernel is taken from git://xenbits.xen.org/people/sstabellini/linux-pvhvm.git. > > > > The most up to date branch is 2.6.35-rc5-pvhvm-v7, please use it for > > your tests. > > Hi, > do you plan to backport your changes on a widely adopted long time > supported 2.6.32.x? > > if you backport the changes, more people could test and bench it on > different hardwares >Yes, you are right. A little while back I backported the whole series to 2.6.32 and the result was included the Jeremy''s pvops main branch, but I haven''t kept it completely up to date. The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards PV spinlocks, the balloon driver and the PV console. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Il 29/09/2010 11:10, Stefano Stabellini ha scritto: [cut]> Yes, you are right. > A little while back I backported the whole series to 2.6.32 and the > result was included the Jeremy''s pvops main branch, but I haven''t kept > it completely up to date. > The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards > PV spinlocks, the balloon driver and the PV console.that''s a good news but in my experience the pv_ops kernel is not as stable as the suse xenified kernel ...imho the best would be to have clean patches for a 3.6.32.x vanilla kernel ...so we can do packages close to the official ones ...with only this great addition. Best regards, Christian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Thu, Sep 30, 2010 at 08:56:08PM +0200, Christian Zoffoli wrote:> Il 29/09/2010 11:10, Stefano Stabellini ha scritto: > [cut] > > Yes, you are right. > > A little while back I backported the whole series to 2.6.32 and the > > result was included the Jeremy''s pvops main branch, but I haven''t kept > > it completely up to date. > > The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards > > PV spinlocks, the balloon driver and the PV console. > > that''s a good news but in my experience the pv_ops kernel is not as > stable as the suse xenified kernel > ...imho the best would be to have clean patches for a 3.6.32.x vanilla > kernel ...so we can do packages close to the official ones ...with only > this great addition. >I think suse xenified kernel already has pv-on-hvm drivers, using the old unmodified_drivers. some info: http://wiki.xensource.com/xenwiki/XenLinuxPVonHVMdrivers -- Pasi _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Il 01/10/2010 08:00, Pasi Kärkkäinen ha scritto: [cut]> I think suse xenified kernel already has pv-on-hvm drivers, > using the old unmodified_drivers.yes but I use a suse-like kernel only on dom0, my domU are mainly debian lenny. Best regards, Christian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Fri, Oct 01, 2010 at 03:19:09PM +0200, Christian Zoffoli wrote:> Il 01/10/2010 08:00, Pasi Kärkkäinen ha scritto: > [cut] > > I think suse xenified kernel already has pv-on-hvm drivers, > > using the old unmodified_drivers. > > > yes but I use a suse-like kernel only on dom0, my domU are mainly debian > lenny. >Debian lenny''s kernel-xen (that you probably use in the domU) is using those suse xenlinux patches. You should be able to build the unmodified_drivers pv-on-hvm for that kernel. Upcoming Debian Squeese (6.0) has a kernel based on upstream pvops, and it should have pv-on-hvm aswell (afaik). -- Pasi _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Thu, 30 Sep 2010, Christian Zoffoli wrote:> Il 29/09/2010 11:10, Stefano Stabellini ha scritto: > [cut] > > Yes, you are right. > > A little while back I backported the whole series to 2.6.32 and the > > result was included the Jeremy''s pvops main branch, but I haven''t kept > > it completely up to date. > > The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards > > PV spinlocks, the balloon driver and the PV console. > > that''s a good news but in my experience the pv_ops kernel is not as > stable as the suse xenified kernel > ...imho the best would be to have clean patches for a 3.6.32.x vanilla > kernel ...so we can do packages close to the official ones ...with only > this great addition. >I have just done a backport to 2.6.32 of the whole series, branch name 2.6.32-pvhvm; I''ll try to keep it up to date. It should be easy to rebase the series on the latest 2.6.32 stable kernel or to apply CVE fixes on top of this branch. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Fri, Oct 01, 2010 at 03:06:12PM +0100, Stefano Stabellini wrote:> On Thu, 30 Sep 2010, Christian Zoffoli wrote: > > Il 29/09/2010 11:10, Stefano Stabellini ha scritto: > > [cut] > > > Yes, you are right. > > > A little while back I backported the whole series to 2.6.32 and the > > > result was included the Jeremy''s pvops main branch, but I haven''t kept > > > it completely up to date. > > > The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards > > > PV spinlocks, the balloon driver and the PV console. > > > > that''s a good news but in my experience the pv_ops kernel is not as > > stable as the suse xenified kernel > > ...imho the best would be to have clean patches for a 3.6.32.x vanilla > > kernel ...so we can do packages close to the official ones ...with only > > this great addition. > > > > I have just done a backport to 2.6.32 of the whole series, branch name > 2.6.32-pvhvm; I''ll try to keep it up to date. > It should be easy to rebase the series on the latest 2.6.32 stable > kernel or to apply CVE fixes on top of this branch. >Hopefully Jeremy merges 2.6.32-pvhvm to xen/stable-2.6.32.x branch.. -- Pasi _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Fri, 1 Oct 2010, Pasi K??rkk??inen wrote:> On Fri, Oct 01, 2010 at 03:06:12PM +0100, Stefano Stabellini wrote: > > On Thu, 30 Sep 2010, Christian Zoffoli wrote: > > > Il 29/09/2010 11:10, Stefano Stabellini ha scritto: > > > [cut] > > > > Yes, you are right. > > > > A little while back I backported the whole series to 2.6.32 and the > > > > result was included the Jeremy''s pvops main branch, but I haven''t kept > > > > it completely up to date. > > > > The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards > > > > PV spinlocks, the balloon driver and the PV console. > > > > > > that''s a good news but in my experience the pv_ops kernel is not as > > > stable as the suse xenified kernel > > > ...imho the best would be to have clean patches for a 3.6.32.x vanilla > > > kernel ...so we can do packages close to the official ones ...with only > > > this great addition. > > > > > > > I have just done a backport to 2.6.32 of the whole series, branch name > > 2.6.32-pvhvm; I''ll try to keep it up to date. > > It should be easy to rebase the series on the latest 2.6.32 stable > > kernel or to apply CVE fixes on top of this branch. > > > > Hopefully Jeremy merges 2.6.32-pvhvm to xen/stable-2.6.32.x branch..The problem is that Jeremy''s stable branch is sufficiently different from stable 2.6.32 that the port in non-trivial. It would probably easier for him to cherry-pick the last 6 patches in the series. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Il 01/10/2010 16:06, Stefano Stabellini ha scritto: [cut]> I have just done a backport to 2.6.32 of the whole series, branch name > 2.6.32-pvhvm; I''ll try to keep it up to date. > It should be easy to rebase the series on the latest 2.6.32 stable > kernel or to apply CVE fixes on top of this branch.that''s great!! ...thank you very much Best regards, Christian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jeremy Fitzhardinge
2010-Oct-01 22:19 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
On 10/01/2010 07:16 AM, Stefano Stabellini wrote:> On Fri, 1 Oct 2010, Pasi K??rkk??inen wrote: >> On Fri, Oct 01, 2010 at 03:06:12PM +0100, Stefano Stabellini wrote: >>> On Thu, 30 Sep 2010, Christian Zoffoli wrote: >>>> Il 29/09/2010 11:10, Stefano Stabellini ha scritto: >>>> [cut] >>>>> Yes, you are right. >>>>> A little while back I backported the whole series to 2.6.32 and the >>>>> result was included the Jeremy''s pvops main branch, but I haven''t kept >>>>> it completely up to date. >>>>> The new changes that are only in the 2.6.35-rc5-pvhvm-v7 branch regards >>>>> PV spinlocks, the balloon driver and the PV console. >>>> that''s a good news but in my experience the pv_ops kernel is not as >>>> stable as the suse xenified kernel >>>> ...imho the best would be to have clean patches for a 3.6.32.x vanilla >>>> kernel ...so we can do packages close to the official ones ...with only >>>> this great addition. >>>> >>> >>> I have just done a backport to 2.6.32 of the whole series, branch name >>> 2.6.32-pvhvm; I''ll try to keep it up to date. >>> It should be easy to rebase the series on the latest 2.6.32 stable >>> kernel or to apply CVE fixes on top of this branch. >>> >> Hopefully Jeremy merges 2.6.32-pvhvm to xen/stable-2.6.32.x branch.. > > The problem is that Jeremy''s stable branch is sufficiently different > from stable 2.6.32 that the port in non-trivial. > It would probably easier for him to cherry-pick the last 6 patches in > the series.I tried this, but it is a bit awkward: Two of them have already been applied verbatim. The blkfront name patch clashes with something similar that''s already in xen/next (the version I currently have does xlbd_reserve_minors(), but this branch''s patch doesn''t). I don''t know which version to take. The hvc console patches clash fairly badly with the dom0 ones; nothing fundamental, but fiddly. The balloon.c changes clash with the largepage ballooning stuff. J _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Fri, 1 Oct 2010, Jeremy Fitzhardinge wrote:> >> Hopefully Jeremy merges 2.6.32-pvhvm to xen/stable-2.6.32.x branch.. > > > > The problem is that Jeremy''s stable branch is sufficiently different > > from stable 2.6.32 that the port in non-trivial. > > It would probably easier for him to cherry-pick the last 6 patches in > > the series. > > I tried this, but it is a bit awkward: > > Two of them have already been applied verbatim. > > The blkfront name patch clashes with something similar that''s already in > xen/next (the version I currently have does xlbd_reserve_minors(), but > this branch''s patch doesn''t). I don''t know which version to take. >just take the latest version> The hvc console patches clash fairly badly with the dom0 ones; nothing > fundamental, but fiddly.skip them for now: they are not critical and at some point I''ll rebase them on top of the "xen initial domain" series that has dom0 hvc_xen support and you''ll be able to pull them without conflicts.> > The balloon.c changes clash with the largepage ballooning stuff. >I have reworked the balloon patch on top of stable 2.6.32, I am attaching the patch to this email. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On Mon, 4 Oct 2010, Stefano Stabellini wrote:> On Fri, 1 Oct 2010, Jeremy Fitzhardinge wrote: > > >> Hopefully Jeremy merges 2.6.32-pvhvm to xen/stable-2.6.32.x branch.. > > > > > > The problem is that Jeremy''s stable branch is sufficiently different > > > from stable 2.6.32 that the port in non-trivial. > > > It would probably easier for him to cherry-pick the last 6 patches in > > > the series. > > > > I tried this, but it is a bit awkward: > > > > Two of them have already been applied verbatim. > > > > The blkfront name patch clashes with something similar that''s already in > > xen/next (the version I currently have does xlbd_reserve_minors(), but > > this branch''s patch doesn''t). I don''t know which version to take. > > > > just take the latest version > > > The hvc console patches clash fairly badly with the dom0 ones; nothing > > fundamental, but fiddly. > > skip them for now: they are not critical and at some point I''ll rebase > them on top of the "xen initial domain" series that has dom0 hvc_xen > support and you''ll be able to pull them without conflicts. > > > > > The balloon.c changes clash with the largepage ballooning stuff. > > > > I have reworked the balloon patch on top of stable 2.6.32, I am attaching > the patch to this email.of course if you would like me to prepare a branch on top of stable-2.6.32.x I can do that as well _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Jeremy Fitzhardinge
2010-Oct-04 19:23 UTC
Re: [Xen-devel] Performance numbers on PV-on-HVM
On 10/04/2010 10:08 AM, Stefano Stabellini wrote:> On Mon, 4 Oct 2010, Stefano Stabellini wrote: >> On Fri, 1 Oct 2010, Jeremy Fitzhardinge wrote: >>>>> Hopefully Jeremy merges 2.6.32-pvhvm to xen/stable-2.6.32.x branch.. >>>> >>>> The problem is that Jeremy''s stable branch is sufficiently different >>>> from stable 2.6.32 that the port in non-trivial. >>>> It would probably easier for him to cherry-pick the last 6 patches in >>>> the series. >>> I tried this, but it is a bit awkward: >>> >>> Two of them have already been applied verbatim. >>> >>> The blkfront name patch clashes with something similar that''s already in >>> xen/next (the version I currently have does xlbd_reserve_minors(), but >>> this branch''s patch doesn''t). I don''t know which version to take. >>> >> just take the latest version >> >>> The hvc console patches clash fairly badly with the dom0 ones; nothing >>> fundamental, but fiddly. >> skip them for now: they are not critical and at some point I''ll rebase >> them on top of the "xen initial domain" series that has dom0 hvc_xen >> support and you''ll be able to pull them without conflicts. >> >>> The balloon.c changes clash with the largepage ballooning stuff. >>> >> >> I have reworked the balloon patch on top of stable 2.6.32, I am attaching >> the patch to this email. > > of course if you would like me to prepare a branch on top of > stable-2.6.32.x I can do that as wellA resolved merge would be better than rebasing onto stable-2.6.32.x. J _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel