- Creating your virtual machine
- Where are the Xen configuration files?
- Where are the Xen log files?
- Troubleshooting common problems
- Was it worth it? Windows performance on Xen
At Tickmeet, our build, test and deployment system for our web apps and back-end services are a one-click affair thanks to the awesome TeamCity. Up until now, our TeamCity server lived on a Rackspace Cloud Server. Not because it makes sense to have a CI server in the cloud (hint: it doesn’t). But simply because we happened to have that server lying around when we decided to give TeamCity a try. And as its happens, our playground server ended up becoming our main (and only) CI server.
Lately however, that setup started showing its limits. Raw performance of cloud servers aren’t great and prices are astronomical for what you get. CI is a terrible use-case scenario for a cloud server.
So we decided to migrate our CI environment to a nifty EX40S-SSD from Hetzner. 32GB of RAM, quad-core processor, a couple of SSDs and great support for less than €50 / month? What's not to love?
Rather than a bare-metal install of Windows, we decided to give Xen on Ubuntu a try so that we could easily move the whole install to a new machine if needed by just taking a snapshot of the VM and moving the image file across. It also was a great opportunity to get up to speed with Linux again after not having used Linux in anger since 2000 (remember when the only way to get hold of a Linux distribution was to get down to your local newsagent and get it on the CD-ROM that came with Linux Magazine?).
What followed was a exercise in pain and frustration. While Linux has definitely made great stride in the last 13 years in usability (I didn't have to recompile the kernel once!), Xen is as lovely to setup today as Linux was 13 years ago.
So hopefully, this little guide will help save a few people from pulling their hair out trying to get Windows up and running on Xen.
I’m assuming that you’ve got a basic understanding of virtualization and Linux and that you’re familiar with using SSH, navigating the file system, running commands as root and editing files with vi. I’ll be using Ubuntu 13.10 and Xen 4.3 on a Hetzner dedicated server.
Before you start: request an additional IP address
You’ll need an IP address for your virtual machine. Unless you already have a spare IP you can use, request one from your host. Some hosts can take a few days to issue an extra address so do that now.
If you’re on Hetzner
- Login to the robot interface, select your server and click the IPs tab.
- Click the “Ordering of additional IP” link at the bottom and order an extra IP stating “Virtual Machine” for the purpose of use.
- It might take a day or two for your IP to get issued.
Xen, xend, xm, xl, XenServer, XAPI, XenAPI, xe, XCP, WTF?
Before jumping in, you should take a minute to get familiar with the terms you’ll be faced with when googling for solutions for your Xen problems (and googling you will). Trust me, it will help. Here we go:
Xen is the hypervisor. It was first created in the late 90’s as part of a research project at Cambridge University by Keir Fraser and Ian Pratt. It was open-sourced in 2002 and is now in version 4.3. Xen powers, among others, Amazon EC2, Rackspace Cloud Server and Linode. If you’re interested in history, check out the History of Xen and Why Xen Project.
xend and xm
Xen itself doesn’t have any user interface - it only expose APIs. Several “toolstacks” have therefore been built to allow humans to create, start, stop, control and otherwise manage virtual machines running on Xen. xend and xm are one of those toolstacks.
xm is a command-line tool that you can use to create and control Xen VMs. xend is a deamon that xm communicates with and that does the actual work.
As of Xen 4.2, xm and xend have been deprecated and replaced with a new toolstack with the lovely name of ‘xl’.
xl is a Xen toolstack introduced with Xen 4.1 and made the default toolstack in Xen 4.2. Just like xm, xl is a command-line tool that you can use to create and control VMs running on the Xen hypervisor. Unlike xm, xl doesn’t rely on the xend deamon to do its work (so you can safely disable xend if you use xl and xend is still running on your server).
xl is the toolstack we’ll use in this guide. xl is mostly backwards-compatible with xm. So if you run into an xm command when googling, simply replace ‘xm’ with ‘xl’ in the command and it should work in the same way.
XAPI, XenAPI, xe
XAPI (short for XenAPI) is another toolstack for Xen, which we won’t be using in this guide. XAPI offers a lot more functionalities than xm and xl.
Unlike xm or xl, XAPI isn’t a command-line tool but instead exposes an XML-RPC interface. You can use XAPI programmatically. Or you can install and use the 'xe' command-line tool, which will call the XAPI for you.
Unlike what its name suggests, XenServer isn’t a part of the Xen hypervisor but is a completely separate product. We won’t be using XenServer in this guide.
XenServer is a “turnkey open source virtualization solution”. It’s essentially a Linux distribution customised for running Xen VMs. It includes Xen (obviously) and the XAPI toolstack. It was initially created by a company called XenSource founded by Ian Pratt, one of the creators of Xen, as a commercial offering for the enterprise market. It was acquired by Citrix in 2007, which kept commercialising and developing it.
Up until recently, XenServer was partly closed-source. In June 2013, Citrix fully open-sourced it.
XCP stands for Xen Cloud Platform. It was the open-source subset of XenServer. Since Citrix fully open-sourced XenServer in June 2013, XCP has effectively lost any purpose and has been discontinued.
Before progressing any further, go and read the Xen Beginner’s Guide. It’s slightly out of date in the details but the main points apply and it’s a very easy to digest introduction to Xen. If you’re feeling lazy, read at least the first section “What is Xen all about?" in order to understand the terms ‘domains’, ‘dom0’, ‘domU’, ‘PV’ and ‘HVM’, which are at the core of everything in Xen.
First things first: pick the right storage medium for your VM - Logical Volume or file-based disk image?
Logical Volumes are the recommended storage medium for best performances of Xen virtual machines. Alternatively, if you haven’t setup LVM on your server, you can use a good old file-based disk image.
If you’ve been living under a rock for the past decade as I have, you’ll have missed the rise of LVM - Linux’ Logical Volume Manager - and you’ll probably be quite confused by now. Think of Logical Volumes as virtual partitions for your drive. Instead of having to decide up-front the size of your partitions, you can just use Logical Volumes instead of partitions. LVM then lets you easily expand or shrink your logical volumes as and when needed. And when I say easily, I mean easily. Check this out:
lvcreate —name my-new-volume —size 60G vg0 lvextend —size 100G /dev/vg0/my-new-volume lvremove /dev/vg0/my-new-volume
It's rare to see such beautifully designed interfaces in Linux. Hats off to LVM’s creator Heinz Mauelshagen for creating such a beauty.
There are many more advantages to using LVM instead of partitions. For a quick overview, check out the Wikipedia article and the “Setup LVM storage for guests” section in the Xen Beginner’s Guide. And if you have time, the HOWTO is also good.
Is your server using LVM?
Find out with:
sudo lsblk -o NAME,FSTYPE,SIZE,MOUNTPOINT
Here’s an example setup:
NAME FSTYPE SIZE MOUNTPOINT sda 223.6G ├─sda1 linux_raid_member 512M │ └─md0 ext3 511.7M /boot └─sda2 linux_raid_member 223.1G └─md1 LVM2_member 223G ├─vg0-swap (dm-0) swap 16G [SWAP] ├─vg0-root (dm-1) ext4 20G / └─vg0-windows (dm-2) 150G sdb 223.6G ├─sdb1 linux_raid_member 512M │ └─md0 ext3 511.7M /boot └─sdb2 linux_raid_member 223.1G └─md1 LVM2_member 223G ├─vg0-swap (dm-0) swap 16G [SWAP] ├─vg0-root (dm-1) ext4 20G / └─vg0-windows (dm-2) 150G
This system is setup with software RAID1, which is why the two physical drives (sda and sdb) are identical.
Our drive is 223.6GB in size.
md0 is a traditional partition of 512MB for /boot (it used to be that /boot couldn’t be on a logical volume, which is why you’ll often see a separate /boot partition even though it seems to be a relic from the past).
The rest of the drive is used up with the md1 partition. As its file system type indicates, md1 is a volume group managed by LVM – i.e. you can create logical volumes on it (and there are 3 logical volumes created on it at the moment with 37GB of unallocated space for you to create more logical volumes).
If you see a logical group on your drive with enough unallocated space to create a virtual machine image (for reference, a base install of Windows Server 2012 R2 Standard with GUI uses up roughly 13GB), you’re good to go!
What if your drive isn’t already setup with LVM?
Solution 1: start from scratch with LVM
Probably the best solution if you’re setting up a new server that’s been setup with normal partitions instead of logical volumes. Just re-format and re-install the OS using logical volumes instead of partitions. Trivial to do if you’re hosting with Hetzner (see below). I’ll let you figure out how to do it with other hosts.
Got a dedicated server from Hetzner?
The bad news: the default Ubuntu install on Hetzner’s servers uses partitions instead of logical volumes. The good news: Hetzner’s installimage utility makes it a 3-click, 5-minute affair to re-install Ubuntu and set it up with LVM:
- Login to the robot interface.
- Select your server, click the Rescue tab and activate the 64-bit Linux rescue system. Make a note of the randomly-generated root password displayed.
- Reboot your server:
sudo reboot(or simply issue an automatic hardware reset via the robot interface)
- SSH into your server as root. You will be greeted by Hetzner’s rescue system. Login with the password noted earlier. Note: SSH will complain that host key verification failed when you try to connect to the rescue system (for good reason since the rescue system is obviously using a different key than the OS that’s installed on your server). Just delete the entry for your server in your local ~/.ssh/known_hosts file and you’ll be good to go.
- Run installimage.
- Select the 64-bit core Ubuntu system
- A config file will then be displayed where you can customise your install. All the defaults are fine. The only part you want to change are the partitioning. This could be a sensible config (read the comments in the config file for an explanation of the syntax):
PART /boot ext3 512M PART lvm vg0 all LV vg0 swap swap swap 16G LV vg0 root / ext4 20G
i.e. a traditional 512MB partition for /boot and everything else setup as a volume group managed by LVM. The volume group contains two logical volumes: one for swap and one for the root file system. 20GB for your root FS is far more than enough if you’ll only be using the server as a Xen host.
If you’ve got enough unallocated space left on your drive or can attach a new drive to your server, just create a volume group there for the purpose of storing your VM image.
Don’t use LVM. A file-based disk image will work fine too, albeit perhaps with reduced performance. We’ll look at how to create a file-based disk image later.
No need to rewrite the wheel, these two short guides will provide you with the basics:
How To Set Up Your Linode For Maximum Awesomeness – Hacker News discussion
My First 5 Minutes On A Server; Or, Essential Security for Linux Servers – Hacker News discussion
It’s a good idea to read the associated Hacker News discussions to get different view points from possibly more seasoned practitioners.
If needed, you can upgrade to the latest Ubuntu distribution with:
(if you’re not sure of the version of Ubuntu you’re using, try
That’s the easy bit:
sudo aptitude install xen-system-amd64
But that’s the AMD package! What if I have an Intel CPU?
The ‘amd64’ in the package name doesn’t refer to AMD CPUs but to the 64-bit version of x86 instruction set, which was initially created by AMD and later adopted by Intel. Wikipedia has a nice overview of the whole thing. So the ’amd64’ package is the correct package for any 64-bit system, including Intel-based systems.
Setup GRUB for Xen
As Xen is a bare-metal hypervisor (you did read the Beginner’s guide, didn’t you?), we need to setup the GRUB bootloader to start Xen before booting Ubuntu.
The configurations for the various systems that GRUB can start are stored in the
/etc/grub.d/ directory. Installing the Xen package will have created a new config in that directory for Xen. All we need to do now is reorder the list of operating systems GRUB can start so that Xen + Ubuntu is at the top of the list and is started by default when the server boots up.
And this is how you do it:
sudo dpkg-divert --divert /etc/grub.d/08_linux_xen --rename /etc/grub.d/20_linux_xen
Once that’s done, you need let GRUB know that its configuration has changed by running:
Create a networking bridge
The fun part. Getting networking to work with Xen VMs is probably what causes most people the most grief. If you’re interested, you can read the Xen Networking guide to learn about the various options available to you. But you don’t have to.
The most commonly used and easiest way to get networking to work with Xen VMs is to create a software bridge on your host machine, which your VM will use to connect to the network. This will result in your host server and your virtual machine appearing as two separate machines on the network – each with their own IP and MAC addresses as if they were two physical machines. Again, check out the Xen Networking guide if you want to know more.
In Ubuntu, the networking interfaces are configured in the
/etc/network/interfaces file. Before doing anything, create a backup copy of this file so that you can revert to the initial networking config should things go wrong:
sudo cp /etc/network/interfaces /etc/network/interfaces.orig
Now edit this file:
sudo vi /etc/network/interfaces
The only thing we’re interested in is the IPv4 config. Everything else can be left untouched.
If your server is setup to use DHCP, you should see something like this:
auto eth0 iface eth0 inet dhcp
eth0 is your physical ethernet interface. If you’ve got more than one ethernet adapter, you might have
eth2, etc configured in there as well. We only need to setup a bridge on the first interface eth0.
The first line indicates that eth0 should be brought up automatically when the system boots up. The second line indicates that the eth0 interface should be configured via DHCP. Simple.
If the server is configured with a static IP address, you’ll have something like:
auto eth0 iface eth0 inet static address 22.214.171.124 broadcast 126.96.36.199 netmask 255.255.255.224 gateway 188.8.131.52
(you might have more or less settings here depending on your config).
What we need to do is create a new networking interface called
xenbr0, indicate that it’s a software bridge that uses the eth0 physical interface and configure it in the exact same way as eth0. Then we need to “unconfigure” eth0 as the bridge interface will now take over. You can theoretically call your new bridge interface whatever you want but the name ‘xenbr0’ is a well-established convention which you would be well advised to follow.
Here we go:
- Edit the eth0 configuration and replace all instances of
- Add the following line to the xenbr0 configuration just after the
bridge_ports eth0(there are more options you could set to configure the software bridge more finely but all the defaults are good. So bridge_ports is the only setting we need to set explicitly)
We now don’t have a configuration for the eth0 interface anymore. So add a new configuration for eth0 and just set it to manual (as the xenbr0 interface is now handling all the networking work):
iface eth0 inet manual
The IPv4 configuration in your /etc/network/interfaces file should now look like this:
If configured via DHCP:
auto eth0 iface eth0 inet manual auto xenbr0 iface xenbr0 inet dhcp bridge_ports eth0
If configured with a static IP address:
auto eth0 iface eth0 inet manual auto xenbr0 iface xenbr0 inet static bridge_ports eth0 address 184.108.40.206 broadcast 220.127.116.11 netmask 255.255.255.224 gateway 18.104.22.168
Set ‘xl’ as the default toolstack
We want to use the latest
xl command-line tool to create and manage our VM (as opposed to the older and now deprecated
xm tool). So we need to set xl as being our default toolstack in the xen config file.
Edit the xen config file located at
sudo vi /etc/default/xen
In that file, set:
Disable saving VM state on dom0 shutdown or store the VM state in a location with enough storage space
By default, Xen automatically saves the state of your virtual machine to disk when your host server shuts down (e.g. when the physical server is rebooted). This is so that it can restore the virtual machine to whatever state is was in when dom0 (your physical server) is back up. This is a very handy feature which can go badly wrong: if you allocate a lot of RAM to your virtual machine, the state of your VM (which is essentially the content of its RAM) can become huge and fill-up your partition.
Whether or not to save VM state and the location where your VM state is going to be saved are defined in the
/etc/default/xendomains config file. So edit this file and either set XENDOMAINS_SAVE to a blank value to disable the feature or make sure that it points to a location with enough storage space.
Now is time to reboot:
Check that Xen has started
If your GRUB config is right, when the server boots up Xen should start in parallel with Ubuntu. When you SSH into the server again, you’ll be in Ubuntu but you can check that Xen has started and is running properly with the following xl command:
sudo xl list
Note that most ‘xl’ commands need to be run as root (hence the sudo).
This command lists all the Xen ‘domains’ (i.e. virtual machines) currently running on the server. Since you haven’t created any virtual machines yet, you will only see the ‘dom0’ domain, which is the host machine (i.e. your physical machine). The output of that command should look like:
Name ID Mem VCPUs State Time(s) Domain-0 0 1022 2 r----- 440.0
If you get an error indicating that Xen isn’t running, go through your GRUB config again. Make sure that you ran the
If you get an error complaining about the toolstack, double-check that ‘xl’ has been set as your toolstack in the
/etc/default/xen config file.
Check that your network bridge is up
You should see your xenbr0 bridge interface there. The output of this command should look like:
bridge name bridge id STP enabled interfaces xenbr0 8000.d43d7ee2ffba no eth0
You can also run
ifconfig and check that the xenbr0 interface has the correct IP address.
Finally, try to ping an external host or wget an external resource to make sure that everything is still working OK on the networking side:
PING google.com (22.214.171.124) 56(84) bytes of data. 64 bytes from fra02s18-in-f9.1e100.net (126.96.36.199): icmp_seq=1 ttl=56 time=6.17 ms
Get the Windows Server 2012 R2 ISO image
Next up, we’ll need the ISO image for the Windows Server 2012 R2 setup. You can download it from MSDN. If you don’t have an MSDN subscription, TechNet has a 180-day trial.
The easiest way to get it on your server is probably to download it on your own machine first and scp it over. If you’re on a Mac or Linux machine:
scp /path/to/windows-server-2012-r2.iso [email protected]:
(don’t forget the colon at the end)
If you’re on Windows, WinSCP works like a charm.
Create a logical volume or a disk image for your virtual machine
If you’ve selected LVM to manage your VM storage
Create a logical volume for your VM. We’ll call it windows-vm and give it a size of 100GB but feel free to call it whatever you fancy and set it to whatever size fits your needs (reminder: a base install of Windows Server 2012 R2 with a GUI uses up roughly 13GB). I’m also assuming that you called your volume group ‘vg0’.
sudo lvcreate -n windows-vm -L 100G vg0
If you prefer to use a file-based disk image
We’ll create a 100GB image file called windows-vm.img. Feel free to name it whatever you fancy and set it to whatever size fits your needs.
Navigate to wherever you want to store the disk image (I personally like to create a ‘xen’ folder in my home directory) and run:
dd if=/dev/zero of=windows-vm.img bs=1024k seek=102400 count=0
Create your virtual machine configuration file
Almost there. All we need now is a configuration file that defines our virtual machine spec: amount of RAM, number of virtual CPUs, location of the disk image, etc.
You can put this file anywhere. I like to create a ‘xen’ directory on my home directory for all my xen configs:
mkdir ~/xen/ cd ~/xen
Now create a file called windows-vm.cfg (or call it whatever you like):
This is what it should contain:
builder='hvm' memory = 24576 vcpus=4 name = "windows-vm” vif = ['mac=00:50:56:00:11:22,bridge=xenbr0'] disk = ['phy:/dev/vg0/windows-vm,hda,w','file:/path/to/windows-server-2012-R2.iso,hdc:cdrom,r'] boot="dc" serial='pty' vnc=1 vnclisten="0.0.0.0" vncpasswd=“somepassword”
This is a minimal file. There are many more settings you could set but all the other settings have fine defaults. Refer to the documentation for the XL Domain Configuration File Syntax to find out what other settings are available.
builder: set to PV by default. But Windows can’t be paravirtualized and needs full virtualisation, hence the HVM.
memory: amount of RAM to allocate to the VM in MB. Set to 24GB here.
vcpus: number of virtual CPUs to allocate to the VM.
name: name of the VM. Set it to whatever you like.
vif: definition of the virtual network interface that will be created for the virtual machine. ‘mac’ is the MAC address that will be allocated to the VM and is optional. If not set, Xen will pick a random one for you. In that case, the MAC might change whenever the VM is shutdown and restarted, which might cause Windows to think that it needs to be re-activated. So hardcoding a MAC address is a good idea. See below for more details on how to pick a MAC address.
disk: definition of the virtual volumes for the VM. Here, we’re defining a hard-drive pointing to the logical volume we created earlier and a CD drive loaded with the Windows Server setup image we downloaded. If you’re using a file-based disk image instead of LVM to store your VM, replace the phy:/ path with:
boot: CD first, then disk.
vnc: Xen includes a built-in VNC server that allows you to view and control your virtual machine via VNC. We’ll need this to install Windows and we’ll then switch to Remote Desktop once Windows has been installed and is connected to the network.
vnclisten: by default, the Xen VNC server only accept connections from the local machine (i.e. your server), which is not going to be very useful. Setting vnclisten to ‘0.0.0.0’ results in connections from any remote machine being accepted.
vncpasswd: anyone who knows the IP address of your server will be able to connect to your virtual machine via VNC while you’re setting it up. So make sure you have a good password here.
Important security note
The VNC protocol is in clear-text and is very simple and well documented. A man-in-the-middle could easily intercept VNC data and view your virtual machine screen as well as all keyboard input. We’ll only be using VNC for the initial setup until we can enable Remote Desktop but the paranoid among you might still want to setup an SSH tunnel to secure VNC data. How to do this is left as an exercise to the reader.
Setting the right MAC address
Your hosting company might only allow machines with known MAC addresses to connect to their network. In this case, you’ll need to either ask your hosting company to provide you with a MAC address to use for your VM or provide them with the MAC address you chose to use.
If you can choose any MAC address, refer to the relevant section in the Xen Networking guide for advice on how to choose correctly.
If you’re hosting with Hetzner
If you’ve requested an additional IP address to use with your VM, Hetzner will also provide you with the MAC address to use with it. You must use the MAC address Hetzner provides or your VM won’t be allowed on their network.
- Login to the robot interface.
- Select your server and click the IPs tab.
- Click the green icon next to the additional IP you were allocated. The MAC address to use with that IP will be displayed. Use that for the
Start your virtual machine
The moment of truth:
sudo xl create ~/xen/windows-vm.cfg
If everything went well, your virtual machine should now be up:
sudo xl list
Name ID Mem VCPUs State Time(s) Domain-0 0 1022 2 r----- 911.1 windows-vm 1 24576 4 r----- 66.3
The ID column in the output of ‘xl list’ is your virtual machine Domain ID. It changes everytime your virtual machine is restarted. When using xl, you can refer to your virtual machine using either its name or its Domain ID.
Check out the troubleshooting section below.
Connect via VNC
Use a VNC client to connect to the IP of your host machine (your host server IP, not your virtual machine IP).
Note that the VNC server implementation in Xen appears to be quite limited and buggy (for those familiar with the VNC protocol, it only supports the Raw and Hextile encodings and seems to have trouble with the initial handshake. If you’re curious, you can check out the source code here).
Mac OS X’s Screen Sharing utility hangs when trying to connect. The RealVNC client for Mac OS X connects successfully but crashes as soon as it tries to display the first frame. I’ve only been able to connect successfully using TightVNC on Windows or Chicken on OS X and only when setting the preferred encoding to either Raw or Hextile. Try Hextile first as it’s much more efficient than Raw but I’ve seen it crash with pseudo-encoding errors. Raw encoding is stable but very slow.
Can’t connect via VNC?
Check out the VNC troubleshooting section below.
You should now see the Windows setup program. Complete the setup. If using a static IP address in your VM, configure the network adapter in Windows with it (configure it as if it was a normal physical server – i.e. using the same mask, gateway server and DNS servers as you’re using on the host machine).
If you’re hosting with Hetzner
You don’t need to manually configure your VM with the additional IP you requested from Hetzner (although you can if you want). Provided that you’ve configured your VM to use the MAC address Hetzner provided for the additional IP address, you can let your VM get its network config via DHCP – it should work.
Once you’ve got the network up and running on Windows, activate Remote Desktop (go to the Local Server section in Server Manager – you’ll see Remote Desktop listed there). You might also need to go to the Windows firewall and enable all the inbound rules for Remote Desktop (public access might be disabled by default).
Can’t get Windows to connect to the network?
Check out the networking troubleshooting section below.
Clean up your virtual machine configuration file
Once Remote Desktop is up and running, edit your VM configuration file:
and delete all VNC entries.
You can also delete the path to the Windows setup ISO image file so that the CD drive configuration looks like
(the first comma isn’t a typo – we’re defining an empty CD drive).
You could also remove the CD drive definition altogether. But leaving your VM configured with an empty CD drive can be handy. With an empty CD drive you can easily attach an ISO image while your VM is running with:
sudo xl cd-insert windows-vm hdc /path/to/an/image.iso
(note that you must specify the absolute path to the ISO image with this command or it will fail with a “received an error message from QMP server” error)
Finally, set the
boot setting to “c” to only boot from disk.
You can now shutdown and restart your VM to start using the new settings:
sudo xl shutdown windows-vm sudo xl create ~/xen/windows-vm.cfg
Configure Xen to start your virtual machine automatically on boot
By default, Xen won’t start your virtual machine automatically on boot. If you want your VM started automatically, place its configuration file in the
/etc/xen/auto/ directory. Or just add a symlink to it:
sudo ln -s ~/xen/windows-vm.cfg /etc/xen/auto/windows-vm.cfg
Alternatively, you can change the path of the directory where Xen looks for virtual machines to auto-start by editing the Xen domains config file at
Install the Windows PV drivers
In Windows, download and install James Harper’s PV drivers – they will give your virtual machine a significant performance boost and are a no brainer.
Fine-tune your Xen install
Now that you’ve got everything up and running, spend a few minutes going through the Xen Best Practices, Tuning Xen for Performance and Roger Pau Monné’s presentation on Xen Performance Tuning.
A lot of the performance tuning tips involve passing command-line parameters to Xen on startup. Many of the example you’ll find on the web explaining how to do this are out-of-date and refer to GRUB1 configuration files which don’t exist anymore in GRUB2.
So here is how to pass command-line parameters to Xen with GRUB2. Here, we’ll restrict dom0 to 1GB of RAM and two virtual CPUs and pin dom0′s virtual CPUs to physical CPUs 0 and 1:
Edit GRUB2′s configuration file:
sudo vi /etc/default/grub
Add the following line to pass the relevant command-line arguments to Xen on startup:
GRUB_CMDLINE_XEN_DEFAULT="dom0_mem=1024M,max:1024M dom0_max_vcpus=2 dom0_vcpus_pin=true"
Let GRUB know that its configuration has been updated:
Curious about the syntax? Want to know what other options Xen can take on startup? Check out the Xen command-line parameters doc.
Keep reading for a handy reference on Xen configuration, logging and troubleshooting and for some benchmarks of our Windows VM.
Where are the Xen configuration files?
Xen uses two configuration files located in
Command-line parameters passed to Xen on startup (if any) will be defined in GRUB2′s configuration file:
(look for the
GRUB_CMDLINE_XEN_DEFAULT setting and remember to run
update-grub if you make any changes to that file)
The network bridge (usually called ‘xenbr0′) that Xen uses to allow virtual machines to connect to the network through the host server’s physical Ethernet adapter is configured in the network interface configuration file
Where are the Xen log files?
Xen writes its logs in a number of different files located in the
/var/log/xen/ directory. You’ll need to be root to be able to navigate to this location.
You can pass the -v option to xl to enable verbose mode.
This tip by Ian Campbell can also come in handy when debugging networking issues:
"Is /etc/xen/scripts/vif-bridge present? If so then a useful debug technique can be to add to the top
exec >>/tmp/hotplug.log 2>&1
which will cause the script output to be logged to /tmp/hotplug.log which might give a clue."
Troubleshooting common problems
xl create fails to create the virtual machine with “unable to add nic devices” or some other crytpic error message
Error messages surfaced by xl are often unrelated to the actual problem.
Your first port of call in case of xl errors should be to check the logs.
Virtual machine creation failure is often due to subtle syntax mistakes in the configuration file. So double-check your virtual machine configuration file.
You can enable verbose mode in xl by using the -v parameter, although the extra information is unlikely to be of much use.
If the problem appears to be related to the xenbr0 network bridge, check out the networking troubleshooting tips below.
My VNC client fails to connect, crashes while trying to display the virtual machine screen or displays a black screen
The built-in VNC server in Xen is temperamental. First, review the VNC section above to make sure that you’re connecting to the right IP address with the right VNC client and the right encoding.
If everything looks in order, check that the VNC server has actually been started on your host server. VNC listens on port 5900 (unless you’ve changed it via the ‘vncdisplay’ option in your virtual machine configuration file). So telnet into it from your server to check that it’s indeed listening:
telnet localhost 5900
You should be greeted by the VNC protocol’s handshake welcome message:
If the telnet connection fails, review your virtual machine configuration file and make sure that you’ve enabled VNC (vnc=1) and that you’re not using a custom port for VNC (there shouldn’t be a ‘vncdisplay’ setting).
If the local telnet connection succeeds, try to telnet from your own machine:
telnet YOUR_SERVER_IP_OR_ADDRESS 5900
If that telnet connection fails, review your virtual machine configuration file and make sure that you’ve configured VNC to allow connection from remote machines (vnclisten=”0.0.0.0″). If you’ve got a firewall between your own machine and your server, make sure that it allows that connection to go through.
The Windows setup program hangs at the blue Windows logo
I sadly don’t have a solution for that one. I ran into this problem during my first attempt to setup a Windows Server 2012 virtual machine with Xen. The setup program would just hang straight away at the splash screen displaying a blue Windows logo on a black background just after the initial “Loading files…” step.
I tried just about everything to get it to work to no avail. I also downloaded several different ISO images from TechNet (Windows 7, 8, Server 2012 and Server 2012 R2 preview) but they all got stuck at the same point, apart from the Windows 7 setup program which worked just fine.
I eventually resigned to the fact that it might be a problem with Ubuntu 13.10 and Xen 4.3, which had been released just a few days earlier. So I started from scratch again with Ubuntu 13.04 and Xen 4.2. The Windows Server 2012 R2 setup program ran just fine on that new setup – no hangs. Before installing Windows, I upgraded to Ubuntu 13.10 and Xen 4.3 to see if I could reproduce the problem. But the setup program ran just fine after the upgrade no matter what I did. I haven’t been able to reproduce the issue.
My virtual machine can’t connect to the network
An entire book wouldn’t be enough to list all the reasons why networking might be broken in a virtual machine. Here are a few starting points.
First: is networking not working at all or are you just not able to connect to external networks? Find out by pinging your host server IP from your virtual machine.
If you can’t ping your host server from your virtual machine, then check that your network bridge has been created correctly. When running ifconfig, you should also see the virtual network interface created by Xen to allow your virtual machine to communicate with the bridge interface (it should be be called vif[DOMAIN_ID]).
You may also want to check the logs (there’s a tip there on adding logging to the shell script that sets up Xen’s virtual network interface).
If you can ping your host server but can’t access external networks, check that you’ve assigned the correct MAC address to your virtual machine. Many hosts will only allow devices with white-listed MAC addresses to connect to their network so check with your host which MAC address you should be using for your virtual machine.
And, of course, if you’ve manually configured the network adapter in Windows, double-check that the mask, gateway and DNS servers are all correct and match what is configured on your host server.
Was it worth it? Windows performance on Xen
Short answer: the raw performance of a Windows virtual machine on Xen is quite disapointing to say the least. In practice however, our TeamCity builds and deployments are 2x to 3x faster than they were on our Rackspace Cloud Server - a big win.
Longer answer: I haven’t run any comprehensive or even vaguely scientific benchmarks. But for one of the key metric that matters for us, disk I/O, performance of Windows Server 2012 R2 on Xen leaves a lot to be desired.
The host server, an EX40-SSD server from Hetzner configured with two SSDs in software RAID1 achieves disk read throughput in excess of 440MB/s in the host Ubuntu system (as measured by dd with 1MB block reads and by the hdparm -t benchmark).
The Windows Server 2012 R2 virtual machine on that same physical machine achieves average disk read throughput of:
- 60MB/s without the PV drivers
- 80MB/s after having installed the PV drivers
- 94MB/s after having applied the tweaks suggested in the Xen Performance Tuning guide
(as measured by HD Tune).
This is a rather measly 21% of the performance of the host machine.
For comparison purposes, a bare-metal install of Windows Server 2008 R2 on an old EX4S server from Hetzner with two 7,200rpm hard drives configured with hardware RAID1 achieves an average disk read of 145MB/s (as measured by the same HD Tune).
A Windows Server 2008 R2 install on a first-gen Rackspace Cloud Server (the one we’re moving away from) achieves an average disk read speed of 118MB/s. That’s even though Rackspace also uses Xen (I was hoping that SSDs would give us a good performance boost but it’s looking like Rackspace have actually done some serious Xen optimization on their side even though the end result is still disappointing).
In practice however, the VM delivers. A TeamCity build, test and deployment cycle for the whole Tickmeet stack used to take a frustrating 5 to 10 minutes on our old Rackspace Cloud Server. This is now down to a much more satisfying two minutes - not even enough time to prepare a cup of coffee.
So was it worth it? Not really from the headache point of view - Xen is anything but enjoyable to work with. It's also interesting to see how critical pieces of functionality in Xen, like the networking setup, are performed by a bunch of shell scripts seemingly cobbled together with sticky tape.
But ending up with a server 3x faster in practice than our old Rackspace server for 1/10th of the price while gaining even more flexibility isn’t too bad.
And that refresher course in Linux sys-admin made it all worth it. While the Tickmeet back-end is mainly built on a .NET stack, projects like PostgreSQL or Go become more appealing by the day. It makes little doubt that Linux is going to start playing a larger and larger role in our tech stack in the coming months. It’s good to have our first Linux server up-and-running.