----- "Dave Augustus" davea@ingraftedsoftware.com wrote:
I finally realized that when running Xen and in Dom0, Xen hides the AMD-V in /proc/cpuinfo
Really?
dom0:
flags : fpu tsc msr pae mce cx8 apic mtrr mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow constant_tsc pni monitor cx16 lahf_lm cmp_legacy svm cr8_legacy
guest:
flags : fpu tsc msr pae mce cx8 apic mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt lm 3dnowext 3dnow constant_tsc pni monitor cx16 lahf_lm cmp_legacy svm cr8_legacy
(...more reasons to move to KVM)
I have a hunch that you are going to do whatever you are told to do and that you really have no idea what is going on.
Can we get more "Hey, I'm told KVM is cool so I'm migrating all of my production servers to it because the free OS of my choice may not support it in a year or two after I leave my current suckjob position and I hear, since I don't read mainline kernel mailing lists religiously and keep only a superficial idea in my head about kernel development, since I really have no idea what 'kernel development' means, that KVM is the only Jesus-approved way of doing things" threads? None of you seem to have have a head for any of this, it seems.
Maybe you're just working for Billy's Interwebs-r-Us and keeping Sally's Nail Shop going strong, but the last time I checked, the "ent" in CentOS stood for "enterprise".
Here are a few tips:
1. Fuck KVM. 2. Stick with Xen because there is quite a lot of time until 5 is EOL'd and if you haven't noticed, it's actually a mature technology. 3. Figure out what you are going to do with Sally's Nail Shop in the meantime. If you have time to fuck everything up in your environment with KVM, you can probably save 20% or more optimizing your environment and even more with proper capital investments and training. 4. Figure out how either a) non-critical/enterprise services would ever be served by KVM's features or lack thereof, or b) you are going to CYA when you can't guarantee an SLA. 5. Spend more time on KVM dev lists instead of posting here and annoying others with your butthurt KVM-won't-work posts, as it's not even supported upstream.
KVM works. I'm happy with it. But then I build servers with 6 guests or less for small businesses.
There are comparisons. They say KVM doesn't scale as well. They say in some areas xen shines, and in some areas kvm shines. But the comparisons are all from last year before red hat released 5.4.
RHEL 5.5, which includes many improvements will be out shortly. Think I'll stick with it...
Hi,
just two questions: 1. Is there anything faster than XEN-paravirtualization?
2. Why XEN 5? XEN 3 is quite stable, too.
I have 31 DomUs up and running on a single Box - and have a strong feeling that even 60 will run flawless. But: All of them are Para-Virtualized.
I have no problem with disk IO-Bottlenecks since my DomUs are not Database-Servers - so there is mostly static information in the filesystems.
I see no reason why I should move to KVM. My only limitation is memory, since RAM is not being virtually mapped yet.
Kind regards
Nils
-----Ursprüngliche Nachricht----- Von: centos-virt-bounces@centos.org [mailto:centos-virt-bounces@centos.org] Im Auftrag von Christopher G. Stach II Gesendet: Freitag, 26. Februar 2010 09:04 An: Discussion about the virtualization on CentOS Betreff: Re: [CentOS-virt] moving from Xen to KVM [...] Here are a few tips:
- F*** KVM.
- Stick with Xen because there is quite a lot of time until
5 is EOL'd and if you haven't noticed, it's actually a mature technology. [...]
On 03/17/2010 02:15 PM, Hildebrand, Nils, 232 wrote:
Hi,
just two questions:
Is there anything faster than XEN-paravirtualization?
Why XEN 5?
XEN 3 is quite stable, too.
I have 31 DomUs up and running on a single Box - and have a strong feeling that even 60 will run flawless. But: All of them are Para-Virtualized.
I have no problem with disk IO-Bottlenecks since my DomUs are not Database-Servers - so there is mostly static information in the filesystems.
The term "paravirtualization" is becoming quite dated. Even if you install a KVM guest without that option if you choose the virtio driver inside then you still end up with "paravirtualized" I/O. With the advent of things like nested page tables and SR-IOV the "fully virtualized=slow, paravirtualized=way faster" logic is no longer necessarily true at least not for every aspect of the system.
Regards, Dennis
(Disclaimer: So far I'm running all my virtualized servers on Xen but use KVM exclusively at home. Eventually I plan to migrate to KVM on the servers too but at the moment I'm not in a hurry but that is not because i think KVM isn't up for the task but because there is no urgent need to do so.)
On Wed, Mar 17, 2010 at 6:35 AM, Dennis J. dennisml@conversis.de wrote:
On 03/17/2010 02:15 PM, Hildebrand, Nils, 232 wrote:
Hi, I have 31 DomUs up and running on a single Box - and have a strong
feeling that even 60 will run flawless.
But: All of them are Para-Virtualized.
I have no problem with disk IO-Bottlenecks since my DomUs are not
Database-Servers - so there is mostly static information in the filesystems.
The term "paravirtualization" is becoming quite dated. Even if you install a KVM guest without that option if you choose the virtio driver inside then you still end up with "paravirtualized" I/O. With the advent of things like nested page tables and SR-IOV the "fully virtualized=slow, paravirtualized=way faster" logic is no longer necessarily true at least not for every aspect of the system.
Regards, Dennis
In the Xen world paravirtualizing will be replaced by Hybrid virtualizing. As hardware virt becomes faster (ie, not so slow) then Xen will change to using HVM as the default and paravirtualize EVERYTHING else. This is not the same thing as KVM which uses hardware virt for cpu, emulation for most things except disk and network which are paravirtualized (if chosen). I look forward to this as HVM in Xen is slower than KVM even though it's kind of doing the same thing. However, I don't think people have benchmarked either enough to realize how much of a hit we're taking with virtio.
KVM has some neat tricks up their sleeve as well like shared memory, nesting etc.. I may put up a KVM box just because I need nesting (for a classroom to teach virtualization).
Grant McWilliams
Some people, when confronted with a problem, think "I know, I'll use Windows." Now they have two problems.
On Wed, Mar 17, 2010 at 03:49:40PM -0700, Grant McWilliams wrote:
On Wed, Mar 17, 2010 at 6:35 AM, Dennis J. <[1]dennisml@conversis.de> wrote:
On 03/17/2010 02:15 PM, Hildebrand, Nils, 232 wrote: > Hi, > I have 31 DomUs up and running on a single Box - and have a strong feeling that even 60 will run flawless. > But: All of them are Para-Virtualized. > > I have no problem with disk IO-Bottlenecks since my DomUs are not Database-Servers - so there is mostly static information in the filesystems. The term "paravirtualization" is becoming quite dated. Even if you install a KVM guest without that option if you choose the virtio driver inside then you still end up with "paravirtualized" I/O. With the advent of things like nested page tables and SR-IOV the "fully virtualized=slow, paravirtualized=way faster" logic is no longer necessarily true at least not for every aspect of the system. Regards,  DennisIn the Xen world paravirtualizing will be replaced by Hybrid virtualizing. As hardware virt becomes faster (ie, not so slow) then Xen will change to using HVM as the default and paravirtualize EVERYTHING else. This is not the same thing as KVM which uses hardware virt for cpu, emulation for most things except disk and network which are paravirtualized (if chosen). I look forward to this as HVM in Xen is slower than KVM even though it's kind of doing the same thing. However, I don't think people have benchmarked either enough to realize how much of a hit we're taking with virtio.
Do you have some benchmarks showing Xen HVM is slower than KVM? I believe so far Xen HVM has actually been faster than KVM.
KVM has some neat tricks up their sleeve as well like shared memory, nesting etc.. I may put up a KVM box just because I need nesting (for a classroom to teach virtualization).
Lately Xen has gotten a couple of different types of memory sharing implementations aswell. At least one of them will ship in upcoming Xen 4.0.0.
-- Pasi
Grant McWilliams
Some people, when confronted with a problem, think "I know, I'll use Windows." Now they have two problems.
References
Visible links
- mailto:dennisml@conversis.de
CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
On Wed, Mar 17, 2010 at 02:15:52PM +0100, Hildebrand, Nils, 232 wrote:
Hi,
just two questions:
Is there anything faster than XEN-paravirtualization?
Why XEN 5?
XEN 3 is quite stable, too.
I guess you mean Citrix XenServer 5.5 with "Xen 5" ?
It's a completely different, full commercially supported product from Citrix, which has the opensource Xen hypervisor as a part of it. Actually it's based on CentOS.
Citrix XenServer 5.x uses Xen 3.x hypervisor.
I have 31 DomUs up and running on a single Box - and have a strong feeling that even 60 will run flawless. But: All of them are Para-Virtualized.
Yeah, Linux PV guests perform and work OK.
-- Pasi
I have no problem with disk IO-Bottlenecks since my DomUs are not Database-Servers - so there is mostly static information in the filesystems.
I see no reason why I should move to KVM. My only limitation is memory, since RAM is not being virtually mapped yet.
Kind regards
Nils
-----Ursprüngliche Nachricht----- Von: centos-virt-bounces@centos.org [mailto:centos-virt-bounces@centos.org] Im Auftrag von Christopher G. Stach II Gesendet: Freitag, 26. Februar 2010 09:04 An: Discussion about the virtualization on CentOS Betreff: Re: [CentOS-virt] moving from Xen to KVM [...] Here are a few tips:
- F*** KVM.
- Stick with Xen because there is quite a lot of time until
5 is EOL'd and if you haven't noticed, it's actually a mature technology. [...]
CentOS-virt mailing list CentOS-virt@centos.org http://lists.centos.org/mailman/listinfo/centos-virt
-
Christopher G. Stach II -
compdoc -
Dennis J. -
Grant McWilliams -
Hildebrand, Nils, 232 -
Pasi Kärkkäinen