Hi Markus,
your emails are rejected by the list because you are sending them from a different email address you are not subscribed with. It sent you an invitation for the other email address.
On Fri, 2015-02-20 at 09:17 +0100, Markus Helmke wrote:
Michel, i'm right with you. I've still some Ubuntu PV working, but everything else I transverred to HVM and I can not mention that this isn't felt slower than PV. On plus - Its easyer to handle things like pci-passthrough. And on top there are spezial drivers, for example for Windows they can be downloaded from Univention, which speeds up Network- and Disk-Performance There are still some Benchmarks where pv is working faster.
Great that you can deliver some facts to my assumptions. The only downside I forgot to mention is that you will need a CPU that supports virtualization. I guess that nowadays nobody is using a processor that doesn't support that because they are slow and consume too much energy.
-Michael
Viele Grüße,
Markus Helmke
Mail: markus@helmke.at mailto:markus@helmke.at
-----Ursprüngliche Nachricht-----
Von:Michael Tremer michael.tremer@ipfire.org Gesendet: Fre 20 Februar 2015 02:05 An: R. W. Rodolico rodo@dailydata.net CC: documentation@lists.ipfire.org >> Wiki Mailingliste documentation@lists.ipfire.org Betreff: Re: Xen Documentation
Hi,
On Tue, 2015-02-17 at 16:48 -0600, R. W. Rodolico wrote:
FYI, a brief (and hopefully accurate) reply to your questions.
paravirtualization (PV) uses the kernel from the underlying DOM0 (the OS running on the bare machine, which runs everything else).
Hardware virtualization (HVM) is completely divorced from the underlying operating system. Instead, it uses its own kernel and some fake hardware exported by the underlying system.
PV's are much faster since one kernel controls everything. The kernel is able to direct resource usage without any intermediate layer. However, you must keep the virtual and the underlying system in sync. One thing I remember is every time you upgrade the kernel on the DOM0, you must copy the new lib's to every virtual. Obviously, you can not run non-Linux systems on a PV. To move a PV from one DOM0 to another, the kernel's and libraries must match, or you need to copy the libraries to the virtual before running it.
HVM's have a layer between them and the underlying kernel. I think that is QEMU, but I don't remember. Since it has to go through a second translation layer, it uses more resources. However, you can have a 2.4 kernel on the DOM0, and be running the latest kernel on the virtual, or even Windows, BSD or OSX. HVM's can be moved between DOM0's with little or no modification (generally a small line in the config file).
All the hypervisors we have base on QEMU. VMware put it in their kernel. VirtualBox is still using userspace but have added lots of graphics features. The Xen project put it in a micro kernel. Of course lots of development has been put in all of these projects and therefore they diverged a lot.
For IPFire, I prefer HVM's because you have spent so much time making sure the kernel and libraries are secure. Thus, even on an older DOM0, I can have a very secure firewall/router. However, some things such as automated shutdown of the IPFire instance are not available since IPFire does not include the HVM device drivers (they use a generic device driver when IPFire is installed). That can cause a problem during server shutdown, since the backup is to "destroy" the virtual, ie pull the plug.
I think that we should have a short mentioning that the HVM version is the preferred one then. It is actually not that much slower any more. CPUs do everything in regards of their jobs. Some devices needs to be emulated which causes a bit more overhead yes. In perspective though IPFire does not read or write a lot of data from/to disk so that there will never be a bottleneck on that end. The virtual network drivers have also a negligible overhead.
The PV approach comes with way more difficulties and I guess that HVM is most used any way.
As far as I know, tying a physical piece of hardware to a single DOMU is still supported, though I have not used that since 3.x, http://wiki.xenproject.org/wiki/XenPCIpassthrough indicates it is still available.
Maybe someone else can contribute this later.
Rod
On 02/17/2015 10:05 AM, Michael Tremer wrote:
Hi,
if I may send you a little wishlist:
There are many many guides about how to set up a Xen server with Debian. I am not sure about the state of all of them. Some of them are based on outdated versions of Debian. I suppose it is best to merge them all together and delete the rest.
http://wiki.ipfire.org/start?do=search&id=xen
Maybe it is a good idea to start a little virtualisation section (http://wiki.ipfire.org/en/virtualization/start) with subsections for Xen (http://wiki.ipfire.org/en/virtualization/xen/start), KVM and all the rest.
Explaining an installation of the IPFire system from scratch should not be necessary but the differences to using Xen should be pointed out somewhere.
I personally am always confused about the HVM and PV thing. What should people use? What are the advantages/disadvantages?
There was also this:
http://planet.ipfire.org/post/dropping-support-for-xen-3-x-deprecation-warni... http://planet.ipfire.org/post/bye-bye-xen-legacy-kernel
Many people use the feature where you can hand over the physical hardware to the guest machine. Does this actually still work? What are the benefits of that?
All this information should be out there somewhere in the wiki. Putting it all together in a nice section with small pages which are easy to find would be really great :)
-Michael
On Tue, 2015-02-17 at 03:10 -0600, R. W. Rodolico wrote:
David,
It would be great if you could get some notes on this. Even if they are rough, it would be better than nothing. Having never messed with KVM (I'm Debian/Xen for the most part), I really don't know much about it, though I do engage in "religious wars" over KVM vs Xen vs Virtual Box with from friends/associates of mine.
Anyway, if you had some brief notes, it would be great to include them.
Rod
On 02/16/2015 11:07 PM, David J. Allen wrote:
On 02/16/2015 08:11 PM, R. W. Rodolico wrote: > I am in the process of creating documentation for a Xen virtual IPFire > install. If anyone else is doing it also, please let me know so we can > collaborate. I'm hopeful to have it complete Friday 20 Feb. > > I am trying an install using the scon image. Assuming that works, should > I also write documentation on how to do it from a standard "installer" > installation? > > Again, if anyone else is doing this, please let me know so we do not > duplicate efforts. > > Rod
I am not writing docs on doing so* but I have been running Ipfire in a VM under KVM on Scientific Linux 6.x for a couple of years now. I think setting up Ipfire was easier than getting the KVM set up right.
The VM host box has 2 onboard NICs, but I'm using a dual-NIC card in the box which is dedicated to the Ipfire VM. The tricky part was figuring out how to give Ipfire exclusive access to the WAN NIC without it being accessible to the KVM host and the other VM guests.
- Although I should do it at least for myself - had to redo my
installation a while back after moving and had left no notes for myself. Which of course resulted in re-inventing my own wheel! ;)
Documentation mailing list Documentation@lists.ipfire.org http://lists.ipfire.org/mailman/listinfo/documentation
Documentation mailing list Documentation@lists.ipfire.org http://lists.ipfire.org/mailman/listinfo/documentation
documentation@lists.ipfire.org
-
Michael Tremer