Paravirtualized CentOS 7 and Ubuntu 14.04 on XenServer

33 Replies to “Paravirtualized CentOS 7 and Ubuntu 14.04 on XenServer”

1. Thanks for the CentOS 7 update!
   Have you tried to install CentOS 7 in XenServer 6.2 using LVM. I cannot get it to work…

   1. From what I understand, there are two conditions that must be satisfied for a system to function under paravirt on XenServer 6.2:

      1. It needs to have a valid grub configuration file in the boot directory or partition that uses a filesystem pygrub can read — I’ve found ext3 to work. So a /boot partition using ext3 is needed if you’re going to use LVM.
      2. The grub configuration can’t use complex GRUB2 features that pygrub can’t understand. CentOS & RHEL 7 use the keywords linux16 and initrd16, which the XenServer bootloader doesn’t understand (until the next release; it’s been fixed in upstream Xen). A kickstart script is the easiest way to automate the installation, most crucially to fudge these grub fixes.

   2. Oh, and by the way, my own benchmarks have shown far better IO and CPU performance when running PVHVM guests than paravirtualized domUs. CentOS itself performed poorly on my test lab, probably because it’s using the oldest kernel of all the major distros.

      If you’re serious about using LVM, the easiest thing to do is use the Other install media option and run CentOS under HVM instead of paravirtualization. You can still benefit from Xen-aware kernel drivers under HVM. There won’t be any need to fudge the bootloader configuration… Just as long as you stick with MBR for now.

      1. Yes, I was using /boot classic partition (ext4), and LVM for / (also ext4). But that doesn’t seems to work.
         I will try with HVM, and see how it works.

         Thank your very much

      2. As I’ve noted above, an ext4 /boot partition will not work. It would need to be ext3.

   3. I’ve done new CentOS 7 PV installation instructions that work. It’s a fully automated install with LVM.

      http://grantmcwilliams.com/item/679-centos7-on-xcp

2. Thanks Frederick, indeed that was the problem. /boot partition must be ext3.
   With CentOS 6 I was using ext4 without problem for the /boot partition, but it doesn’t work for 7.

   LVM is fine for PV or HVM.

   With PV machines netperf is showing better numbers than with CentOS 6.5 (8138.38MB v6.5 vs 11620.80MB v7.0; local machine test).

3. Sorry I am a newbie at this. I did the install and found the user centos and pw of Asdfqwerty. How can I find out what the root password is, or can you advise me of the password.

   1. Seems you are quite new to Linux! No worries.

      The root password is not set by default. Instead, you have a user (centos or ubuntu) which has escalation privileges. That is, the user is allowed to use sudo to do tasks as if the user is root. Each time* you use sudo, you will be asked for the current user’s password, not the root password. (* there’s a timeout built in so that you won’t be asked again if you recently authenticated) That’s also why you should use passwd to change your password after installing the image, or create a new user with your own password and delete the default user.

      All this also means you get to:

      1. sudo passwd to set the root password, so that you can log in as root
      2. sudo su to become root and get a root-privileged terminal using only your non-root (centos or ubuntu) password

      The VMs in this post are configured not to have a root password so as to protect against malicious hackers’ attempts to SSH in as root.

      1. @ Frederick. Thank you for the quick response. I really do appreciate your help. Now on my way to get that new car smell away on Linux. :)

      2. Hey @relevelence, you should consider using the new PVHVM images posted at http://www.frederickding.com/posts/2014/07/pvhvm-centos-7-xenserver-312113/ . CentOS 7 is really slow in full paravirtualization.

4. Pingback: PVHVM CentOS 7 on XenServer « Frederick's Timelog

5. Fredrick, Great article! I’ve done a fair bit of CentOS6 on XenServer, but not with kickstart. I used your kickstart script because I was not having success with my previous method. Install finished, but it still failed to boot with a pygrub RuntimeError “line 850” “Unable to find partition containing kernel”. I did make a few edits to the script, mainly to network specifics. Do we need to be on XenServer 6.2? Thanks

   1. Yes, I can only guarantee these scripts on XenServer 6.2 and Creedence. Older versions may not be able to boot from certain partition types. You’ll need a separate /boot partition on ext2 if you are using older XenServer (change the line that makes the /boot partition to use ext2 instead of ext3).

      1. That’s what I suspected. Thanks

6. Just a quick note to say your kickstart script was a godsend! I struggled to get centos 7 working in Xen all day until I came across your pygrub fix! Many thanks.

7. This is great but I need to install Gnome and as sudo I cannot do this. Is there a workaround for this?

   1. What do you mean? You can install GNOME under sudo with a package manager. In CentOS:

      sudo yum groupinstall "GNOME Desktop"

8. XenServer Creedence versions after alpha 4 have removed the two Ubuntu 14.04 PV templates and replaced them with a single Ubuntu 14.04 which seems to install an HVM machine, much like installing with the “Other install media” option. The current Creedence version is beta 2, which I am using right now.

   What would the process be for this scenario?

   Would choosing “Install a minimal virtual machine” from the [F4] menu in the installer’s initial menu be the way forward, here?

   I am going to try this, but would like to know if there is a way to tell, post installation, just what type of machine I have installed; PV or HVM. This is what annoys me particularly about XenServer and XenCenter: There is no obvious way to tell, or to specify this distinction upon installing.

   *TL;DR*

   The main reason for this even bothering me is that I have found that PCI devices like storage controllers are passed through differently on PV vs. HVM, with some working on one and not the other. Furthermore, I have noticed that multiple device pass-through via the XAPI ‘xe’ command works on HVM and not PV, even though hot-plugging via the ‘xl’ command works for multiple devices, regardless.

   Examples from my notes:

   # Passthrough “00:11.0 SATA controller: Advanced Micro Devices, Inc. [AMD/ATI] SB7x0/SB8x0/SB9x0 SATA Controller [AHCI mode] (rev 40)” #
   # Passthrough “01:00.0 SATA controller: Marvell Technology Group Ltd. 88SE9230 PCIe SATA 6Gb/s Controller (rev 10)” #
   # Passthrough “05:00.0 RAID bus controller: Silicon Image, Inc. SiI 3132 Serial ATA Raid II Controller (rev 01)” #
   # Passthrough “09:00.0 Multimedia controller: Philips Semiconductors SAA7160 (rev 02)” #

   [root@Server ~]# xe vm-param-set other-config:pci=0/0000:00:11.0,0/0000:01:00.0,0/0000:05:00.0,0/0000:09:00.0 uuid=2ec3b16c-5cf7-23a6-5f92-5401b6024166
   [root@Server ~]# xe vm-param-list uuid=2ec3b16c-5cf7-23a6-5f92-5401b6024166 | grep other-config

   other-config (MRW): pci: 0/0000:00:11.0,0/0000:01:00.0,0/0000:05:00.0,0/0000:09:00.0;
   vgpu_pci: 0/0000:07:00.0,1/0000:07:00.1; base_template_name: Other install media; mac_seed: 49735cec-8980-eb09-97d4-05c1ca1ebde7;
   install-methods: cdrom

   # Passthrough “09:00.0 Multimedia controller: Philips Semiconductors SAA7160 (rev 02)” #

   [root@Server ~]# xl list

   Name ID Mem VCPUsStateTime(s)
   Domain-0 0 2046 4 r—– 250.8
   Ubuntu Trusty Tahr 14.04 (64-bit) (1) 10 2048 2 -b—- 5.9

   http://blog.xen.org/index.php/2012/06/04/xen-4-2-preview-xl-and-pci-pass-through/

   NB: Devices will only become available after you have previously passed them through using
   xe vm-param-set other-config:pci=

   [root@Server ~]# xl pci-assignable-list

   0000:00:11.0
   0000:07:00.0
   0000:07:00.1
   0000:09:00.0

   [root@Server ~]# xl pci-attach 10 09:00.0 1

9. Thanks for your post, it was very helpful.

   While trying to install xen-tools, I got this error:

   [centos@localhost Linux]$ ./install.sh
   Fatal Error: Failed to determine Linux distribution and version

   I’ve installed manually with the following steps:

   1) Mount dvd
   sudo mount -r -t iso9660 /dev/xvdd /mnt/dvd

   2) Install xentools
   sudo rpm -i /mnt/dvd/Linux/xe-guest-utilities-6.2.0-1120.x86_64.rpm

   3) Reboot
   sudo reboot

   It worked!

10. Great article, thanks. How is possible to understand, when I use PV and when HVM mode? Is this possible by using PV-bootloader=pygrub and HVM-boot-policy=BIOS\ order?
    Thanks

    1. You can determine the type of virtualization using the flags you just mentioned. If PV-bootloader is set to pygrub, you’re certainly using paravirtualization; if HVM-boot-policy is set to that, you’re probably using HVM. You can also check from inside the VM:

       • Execute lscpu. If it shows Virtualization type: full, you’re using (PV)HVM. If it shows Virtualization type: para, you’re using PV.
       • See this answer on StackExchange; basically, try executing dmidecode -s system-product-name. If you are on (PV)HVM, it will display HVM domU.

       PV or PVHVM: which to choose?

       Whether you choose to deploy PV or PVHVM depends on (a) the physical hardware you’re using, (b) the type of workload (e.g. I/O-limited, or CPU-limited) you’re running within the VMs, and (c) the kernel/software used within the VM. I ran some primitive benchmarks comparing CentOS 7 under PV and PVHVM on a dual-CPU Xeon system and comparing Fedora 20 under PV and PVHVM, showing that, in general, you’ll get better CPU performance under PVHVM.

       

       Converting from one to the other

       Note that the images I’ve prepared for PV lack packages that may be necessary for proper function under PVHVM. See my PVHVM guide instead. PVHVM -> PV is an easier conversion than PV -> PVHVM.

11. Fatal Error: Failed to determine Linux distribution and version

    Install Xentools manually
    1. Attach the xs-tools.iso
    1. $ sudo mount /dev/xvdd /mnt
    2. $ sudo yum localinstall /mnt/Linux/xe-guest-utilities-*_64.rpm
    3. $ sudo umount /mnt
    4. Detach the xs-tools.iso

12. Hi Frederick,

    I’m stuck with an Ubuntu installation.

    In the first place, I tried to install Ubuntu 14.04 from iso image on Xenserver 6.5 with no success, though it should be supported in this XenServer version and a dedicated template is present. Unfortunately I cannot use the network install with kickstart because of some network configuration constraint. So I downloaded your prebuilt image, extracted and imported. The import command seems to work and returns the uuid of the task but the task then stays “pending” at 0% progress (it’s about one hour now)… Some hint?

    Thank you

    S

13. It sounds like you’re working on the command line, right? xe vm-import should not return the UUID of a task. It should return the UUID of the VM or template that was successfully imported.

    Here’s what that process should have looked like:

    1. Decompress the XZ archive.

    [root@lightning templates]# xz -d ubuntu-14.04-20140720-pvhvm-template.xva.xz
    

    2. Import with xe vm-import.

    [root@lightning templates]# xe vm-import filename=ubuntu-14.04-20140720-pvhvm-template.xva
    5032a3b3-15a1-fefa-6c34-4eac9ddfb8b4
    

    The UUID returned should be randomly generated.

    3. Verify that the template exists.

    [root@lightning templates]# xe template-list uuid=5032a3b3-15a1-fefa-6c34-4eac9ddfb8b4
    uuid ( RO)                : 5032a3b3-15a1-fefa-6c34-4eac9ddfb8b4
              name-label ( RW): Ubuntu 14.04 PVHVM Template
        name-description ( RW): 
    
    

    In XenCenter, you’ll see the template as well.

    

    In any case, you should probably avoid using PV guests at this point. PVHVM will be better.

14. Pingback: Using OpenStack images on XenServer - Fedora 22, CentOS 7 « Frederick's Timelog

15. OK! can you lemme know now how can I increase the HDD of VM? All tutorials on web

16. Hi, can you tell us the root password you set for Centos ?

17. Sorry, i’ve just forget to read the response already given.

18. I am trying to use your example to rename the virtual machine /etc/hostname
    when does that script run, during install or are we supposed to run it manually afterwards?

19. If you’re using this script to name the VM using its XenServer hostname, it is not executed automatically per the kickstart above. That’s because it depends on having XenServer Tools installed, to retrieve the data from the virtualization host.

    However, if you are simply trying to set the hostname, you should be able to do that in the post-install script within the kickstart. Instead of downloading that script, just echo your desired FQDN to /etc/hostname (e.g. echo glorious.example.com > /etc/hostname).

    It is worth noting that CentOS 7 should have the hostnamectl utility. After the machine has been booted, it seems to be preferable to use the hostnamectl utility to set the instance hostname.

20. This article was really helpful for us resolving some issues with PV 14.04 on XenServer 6.2.

    Now i try to run Ubuntu Server 16.04 on XenServer (still) 6.2. Have you tried that?

    The VM is booting but i get no image in XenCenter. Maybe due hvc0 is not available.

    I also created a bug report: https://bugs.launchpad.net/ubuntu/+source/systemd/+bug/1587009

    1. So what PV does is send the TTY to a virtualized serial port that XenServer eventually puts on the management console. In previous kernels/distros, that device shows up as /dev/hvc0 inside the guest. (FYI, on the dom0, you can use xl console [vm friendly name] to get to the same thing.)

       I recently tested a CentOS 7 HVM installation and tried to output to that same serial console (yes, you can do this from inside a HVM guest too). It didn’t work. Through trial and error, I discovered that the virtualized serial port is now actually /dev/ttyS0 … so if you’ve had trouble using hvc0, try ttyS0.

Comments are closed.