1240
Comment:
|
2860
Xen 3.2 vs. KVM 60
|
Deletions are marked like this. | Additions are marked like this. |
Line 1: | Line 1: |
While benchmarks are a very difficult exercice and as such the results must be considered carefully, they are still interesting to compare the respective performance between the solutions for different workloads. This page will try to keep an eye on the relative performances of Xen (opensource version) and KVM for a non paravirtualized CPU (i-e with HVM). | While benchmarks are a very difficult exercixe and as such the results must be considered carefully, they are still interesting to compare the respective performance between the solutions for different workloads. |
Line 3: | Line 3: |
For the following 2 tests, the system is an intel core 2 duo 2.4Ghz w/ 2GB RAM running the latest available versions: | This page will try to keep an eye on the relative performances of Xen and KVM. Until virtio is not part of the mainstream linux kernel, the tests will concentrate on performance on HVM systems, so the guest system will be an "unmodified OS". |
Line 5: | Line 5: |
* for xen: Xen 3.1.0 + Dom0 kernel based on kernel.org 2.6.18 | = 12/02/2008 = Thanks for Dan Berrange who pointed me on xen.org instead of xensource.com as a source of download for the latest Xen. The tests consists of the compilation of a linux kernel 2.6.24 with allnoconfig (make allnoconfig && time make). The host hardware is a 1.86Ghz Intel Core2duo. == unmodified VM: 64bits Debian Lenny == The first two tests run the same Virtual Machine with an unmodified 64bits Debian Lenny. * Host is an opensource Xen-3.2 (from xen.org) with a Dom0 running a kernel-2.6.18-xen, Dom0 userspace is a 64bits Debian Lenny || '''run''' || '''1''' || '''2''' || '''3''' || || '''real''' || 2m26 || 2m19 || 2m20 || || '''user''' ||1m21 || 1m22 || 1m21 || || '''sys''' || 1m02 || 0m58 || 1m00 || * Host is a 64bits Debian Lenny with a 2.6.24 kernel and KVM-60 || '''run''' || '''1''' || '''2''' || '''3''' || || '''real''' || 2m11 || 2m07 || 2m07 || || '''user''' || 1m26 || 1m26 || 1m26 || || '''sys''' || 0m45 || 0m40 || 0m40 || == paravirtualised VM: 64bits Debian Lenny with xenified kernel == The following test runs on the same hardware than the previous one. * Host is an opensource Xen 3.2 (from xen.org) with a Dom0 running a kernel-2.6.18-xen, Dom0 userspace is a 64bits Debian Lenny || '''run''' || '''1''' || '''2''' || '''3''' || || '''real''' || 1m34 || 1m33 || 1m33 || || '''user''' || 1m17 || 1m18 || 1m18 || || '''sys''' || 0m15 || 0m15 || 0m15 || = 11/02/2008 = For the following 2 tests, the system is an Intel core 2 duo 2.4Ghz w/ 2GB RAM running the latest available versions: * for Xen: opensource Xen 3.1.0 (from xen.xensource.com) + Dom0 kernel based on kernel.org 2.6.18 |
Line 8: | Line 47: |
The guest used in both case is the exact same VM (64 bit Debian etch with 256 MB RAM). | The guest used in both case is the exact same VM (non-paravirtualised 64 bit Debian etch with 256 MB RAM). |
Line 10: | Line 49: |
= Guest system boot-up time = | == Guest system boot-up time == |
Line 16: | Line 55: |
= Kernel compilation benchmark = | == Kernel compilation benchmark == |
While benchmarks are a very difficult exercixe and as such the results must be considered carefully, they are still interesting to compare the respective performance between the solutions for different workloads.
This page will try to keep an eye on the relative performances of Xen and KVM. Until virtio is not part of the mainstream linux kernel, the tests will concentrate on performance on HVM systems, so the guest system will be an "unmodified OS".
12/02/2008
Thanks for Dan Berrange who pointed me on xen.org instead of xensource.com as a source of download for the latest Xen.
The tests consists of the compilation of a linux kernel 2.6.24 with allnoconfig (make allnoconfig && time make). The host hardware is a 1.86Ghz Intel Core2duo.
unmodified VM: 64bits Debian Lenny
The first two tests run the same Virtual Machine with an unmodified 64bits Debian Lenny.
- Host is an opensource Xen-3.2 (from xen.org) with a Dom0 running a kernel-2.6.18-xen, Dom0 userspace is a 64bits Debian Lenny
run |
1 |
2 |
3 |
real |
2m26 |
2m19 |
2m20 |
user |
1m21 |
1m22 |
1m21 |
sys |
1m02 |
0m58 |
1m00 |
- Host is a 64bits Debian Lenny with a 2.6.24 kernel and KVM-60
run |
1 |
2 |
3 |
real |
2m11 |
2m07 |
2m07 |
user |
1m26 |
1m26 |
1m26 |
sys |
0m45 |
0m40 |
0m40 |
paravirtualised VM: 64bits Debian Lenny with xenified kernel
The following test runs on the same hardware than the previous one.
- Host is an opensource Xen 3.2 (from xen.org) with a Dom0 running a kernel-2.6.18-xen, Dom0 userspace is a 64bits Debian Lenny
run |
1 |
2 |
3 |
real |
1m34 |
1m33 |
1m33 |
user |
1m17 |
1m18 |
1m18 |
sys |
0m15 |
0m15 |
0m15 |
11/02/2008
For the following 2 tests, the system is an Intel core 2 duo 2.4Ghz w/ 2GB RAM running the latest available versions:
- for Xen: opensource Xen 3.1.0 (from xen.xensource.com) + Dom0 kernel based on kernel.org 2.6.18
- for KVM: kvm-60 on a 2.6.23
The guest used in both case is the exact same VM (non-paravirtualised 64 bit Debian etch with 256 MB RAM).
Guest system boot-up time
plateform |
time in seconds |
Xen |
23 |
KVM |
20 |
Kernel compilation benchmark
Time to compile a 2.6.24 allnoconfig
- first try - cold cache
plateform |
real time |
user time |
sys time |
Xen |
1m53 |
1m06 |
0m43 |
KVM |
1m03 |
1m06 |
0m33 |
- second try - w/ cache
plateform |
real time |
user time |
sys time |
Xen |
1m52 |
1m07 |
0m43 |
KVM |
0m56 |
1m07 |
0m31 |