KVM 仮想化へようこそ - 徹底した紹介

私の仮想化セクションを読んでいるなら、私がこれまで主に VMware と VirtualBox に焦点を当てており、クラウド関連とイメージのリマスタリングが少しあったことをご存知でしょう。さて、分岐する時間です。今日、KVM を使用したまったく新しい時代のチュートリアルを正式に開始したいと思います。後で、Xen やその他の奇妙な獣が出てきますが、今のところ、私たちのトピックは KVM です。

このガイドから始めて、RedHat によって開発された KVM (カーネルベースの仮想マシン) テクノロジについて学びます。これは、他の商用ソリューションに代わる無料のオープンソースの代替手段として利用できます。 KVM のダウンロード、インストール、セットアップの方法、仮想マシンの管理に使用できるツール、利用可能なオプション、コマンドラインの使用法とスクリプト作成などについて学習します。また、高度なネットワークとストレージの構成、およびその他の優れた機能についても説明します。それでは始めましょう。

KVM - 用語集

KVM がどのように機能するかについての非常に簡単な概要から始めましょう。重要なことは何もありません。基本的な用語を理解できるように、ほんの少しだけ紹介します。心配はいりません。すぐにすべてがとても簡単になります。

KVM は、Intel-VT および AMD-V として知られる最新の Intel および AMD プロセッサの CPU 仮想化テクノロジ拡張機能を利用することによって機能します。メモリにロードされたカーネル モジュールを使用して、KVM はプロセッサを利用し、変更された QEMU に基づくユーザー モード ドライバーを介して、仮想マシンを作成および実行できるハードウェア レイヤーをエミュレートします。 KVM は CPU 拡張なしでも実行できますが、QEMU を使用して純粋なエミュレーション モードで実行されるため、パフォーマンスが大幅に低下します。

KVM は、VMware 製品や VirtualBox と同様のグラフィカル管理ツールを介して、またはいくつかの方法を使用してコマンド ラインを介して管理できます。

最も一般的な GUI は、RedHat によって開発された Virtual Machine Manager (VMM) と呼ばれます。このツールは、virt-manager という一般的なパッケージ名でも知られています。これには、virt-install、virt-clone、virt-image、および virt-viewer などの多数のサポート ツールが付属しており、それぞれ仮想マシンのプロビジョニング、クローン作成、インストール、および表示に使用されます。 VMM は Xen マシンもサポートしています。

一般的な KVM コマンド インターフェイスは、virsh によって提供されます。具体的には、仮想マシンを作成するための virt-install などのサポート ツールを使用できます。 Ubuntu には、Canonical によって開発された、Ubuntu ビルドのプロビジョニングに使用できる特別な ubuntu-vm-builder ツールがあります。

たくさんの名前が紛らわしいかもしれませんが、簡単に考えてみてください。ただし、もう少し混乱させるために、KVM には QEMU に似た独自の構文もあります。これは、仮想マシンを管理するための推奨される方法ではありません。

詳細については、次のページを参照してください:

カーネル仮想マシン - コミュニティ Ubuntu ドキュメント

howtoforge.com の KVM チュートリアル

KVM/Virsh - コミュニティ Ubuntu ドキュメント

よし、それがステップ 1 だった。

KVM - 長所と短所

KVM を使用する必要がありますか?まあ、それはあなたが何を達成したいかによります。

あなたがカジュアルなホーム ユーザー、仮想化愛好家、または中級者の場合、KVM を使い始めるのは少し難しいかもしれません。これは主にコマンドライン ツールであり、VMware や Virtualbox 製品ほど直感的でもフレンドリーでもありません。 KVM は、GUI 管理に関しては競合他社に何年も遅れをとっているように感じますが、配信に関しては同様に強力です。実際、コマンドライン ツールとして使用するように設計されているため、長期的には生産性と制御性が向上します。そのため、KVM はビジネス環境に最も適しています。

さらに、CPU が仮想化をサポートしていない場合、KVM は非常に遅く非効率的なエミュレーション モードで実行されるため、時間の無駄になります。 KVM は Virtualbox と競合することも知られていますが、それについては別のチュートリアルで詳しく説明します。

したがって、KVM は、仮想化に強い関心を持っている人、おそらく管理者や専門家向けです。それはあなたの最も抱きしめたくなる家庭用おもちゃではなく、今後何年もそうではないかもしれません.しかし、掘り下げていく気があるなら、完全に新しい言語をマスターすることができ、それは後で他のテクノロジーとの力を倍増させるのに役立ちます. VMware や Virtualbox は GUI ツールを使用することを好みますが、KVM はコマンドライン スクリプトの本来の効率性を重視しています。

要約すると、KVM は、使用モードが無制限で、追加のライセンス料や機能の階層化がなく、強力なコマンド ライン インターフェイスを備え、手を汚すことを恐れない、無料の最新の仮想化ソリューションを探している場合に適しています。 CPU に仮想化拡張機能がなく、スクリプトを書くのが怖く、コードに手を出したくなく、VMware Server、ESXi、Virtualbox などの管理が簡単なソリューションを好む場合、KVM は適していません。

よし、行こう。

Test platform

You can use KVM in pretty much any Linux distro. In general, RedHat flavors favor KVM. RHEL ships with KVM enabled in the operating system, so you can also find it in RedHat-based distros like CentOS, Scientific Linux, or Fedora.

Since I'm running mostly Ubuntu at home, I decided to test in Ubuntu, installed on my rather brand-newish HP laptop, which has a revvy i5 processor that supports virtualization. Later this year, I'm planning on deploying either the newest release of CentOS or Scientific Linux on my Linux-dedicated multi-boot machine, so you'll see additional tutorials for those distros, too. We'll get to Xen there, too.

This tutorial shows how to setup KVM on a 64-bit Ubuntu Lucid, the Long Term Release. Sequel guides and howtos will cover many other topics related to KVM, including using it alongside other virtualization technologies and resolving conflicts, advanced management functions, scripting, and more.

Now, finally, let's begin in earnest.

Prerequisites

You need a CPU that supports virtualization. This can be checked thusly:

egrep -c '(vmx|svm)' /proc/cpuinfo

If you get a non-zero value, you're good. BUT ... you will also need to check that virtualization technology is enabled in BIOS. Indeed, after enabling the feature, you will have to cold power-cycle the machine for the change to take effect.

Once this is done, boot into your operating system and get KVM installed.

To check, please run kvm-ok:

Download &install KVM

To get KVM install, please get these packages (true for apt commands):

apt-get install qemu-kvm libvirt-bin  

Optionally, you can install these - most recommended:

apt-get install bridge-utils virt-manager python-virtinst

追記The package names will vary between distributions. For instance, virt-install may come branded as python-virt-install or python-virtinst or even virt-install. The dependencies for virt-clone, virt-image and virt-viewer should be automatically resolved.

Contrary to what most tutorials mention, you do NOT need bridge utilities. You only need them if you want to bridge your virtual adapters with your physical cards. I've shown you a detailed example of this in my Virtualbox networking tutorial. The principle is the same.

Most tutorials also mention the fact that most Wireless network interfaces do not support bridging. This may be the case, but in all my testing, I was able to successfully bridge my Wireless adapters, so keep your hopes high.

VMM (virt-manager) is highly recommended. Likewise, you want all the helper tools, including virt-viewer, virt-install, virt-image, and virt-clone.

Lastly, you can also opt for ubuntu-vm-builder:

apt-get install ubuntu-vm-builder

You will also get a whole bunch of dependencies auto-resolved, so your downloads can take a while.追記On RedHat, use yum install. On SUSE, use zypper install.

Conflict with Virtualbox

Again, contrary to what most tutorials say:you can have both KVM and Virtualbox installed on the same host. You may not get them running both at the same time. In other words, one or the other kernel module needs to be unloaded from memory. But there's no reason to rush ahead. Try, see if it works for you, and if not, only then take corrective actions.

But this can be done in-vivo, so no worries there. You won't need to reboot or anything as drastic as that. I'll show you a separate tutorial that handles this case. In fact, I have both installed and running on the test host.

Using KVM

Now, the really fun part. We will start with the VMM, as it is a GUI tool and this is what most people will and should start with. For all practical purposes, it's very similar to other tools you know, like the VMware console or perhaps Virtualbox.

Virtual Machine Manager (VMM)

Let's take a tour.

The first time you launch the program, it will show two categories, both not connected. These refer to your localhost KVM modules, still not in use. To use, right-click and choose connect.

To add new connections, click File> Add Connection. This will open a window that lets you specify the Hypervisor type and the connection type. VMM can use both local and remote connections, including QEMU/KVM and Xen. You can connect using all kinds of authentications methods.

You can also autoconnect, which means that the next time you fire up VMM, these connections will be ready for you. This is similar to what VMware Server asks you when you start the program. Just for reference:

Kernel versus Usermode

You may be asking what the difference between normal/default and Usermode is? Well, Usermode allows you to run without administrative privileges, but your network capability will be limited. All right, let's move on.

VMM overview, continued

Let's see other functions, before we create a virtual machine.

Network functionality can be reviewed or changed under Host Details. We will have a separate tutorial that teaches how to work with KVM network. And we'll setup bridge utilities, too, so don't worry.

Likewise, you can change the storage settings. There'll be a dedicated guide, I promise:

Change preferences

The VMM has a handful of options you can tweak to make it look and behave smarter. You can enable the system tray icon, set a statistics polling interval, enable polling for disk and network metrics, which are deselected by default, configure console keyboard grabbing, console scaling, audio options, and set confirmation windows for machine activities.

Later on, all of these will allow us to view our virtual machine metrics in more detail and with better control. For instance, showing an Ubuntu guest CPU, Disk I/O and Network I/O:

Here's the system tray icon:

Finally, we're ready to create a new virtual machine.

Create virtual machine

We can do this using VMM or the command line. We will begin with the GUI. The first step is quite intuitive. Enter the name and the install media location. You can use local storage in the form of ISO images or CD/DVDs, HTTP or FTP servers, NFS area, or PXE.

We will use local install media. Next, choose either a physical device or an image. In our case, we will use an ISO. Then, choose the OS type and the version. This does not have to be precise, but an educated guess is good. It allows additional virtual machine optimization.

Allocate RAM and CPUs.

The fourth step is quite interesting - storage. You can create a new image or use an existing one. You can set the size and preallocate all space, which is good for performance, plus it reduces fragmentation.

We will focus soon quite a bit more on the storage. For now, let's either assume we have a disk ready - or that we will create one in the default storage pool. Please note that if you're running in Usermode, you won't be able to write to /var, which is the default storage pool location. You will need to create a new storage pool that is accessible by non-admin users, but more about that in a separate article.

Step 5 is a summary, with some advanced options. You can change your network type, set a fixed Mac address, choose the virtualization type, and the target architecture. If you're running in Usermode, your network options will be limited. If you're not using bridge utilities, you won't be able to bridge your interfaces. Lastly, if you have no CPU virtualization extensions, the Virt Type:field will read QEMU and not KVM. We will see an example of this later on, when we discuss the major drawbacks of running in the emulation mode. For now, here's what a typical Ubuntu virtual machine setup looks like:

Our machine is ready to use. We will soon power it up and use it.

Virtual machine options

Like VMM itself, the VM console has a few interesting options. You can send signals to your guest, you can switch virtual consoles, force reboot and power off, clone, migrate, save, take screenshots, and more. Again, very similar to the competition.

Here's a pair of screenshots teasers for the clone and migrate options. Down the road, we will have detailed tutorials, focusing both on the GUI and CLI functionality.

Virtual machines running

The fun part. Here's a handful of nice screenshots ...

Let's begin with Ubuntu 10.10 Maverick, 32-bit version, boot menu:

Puppy Linux, shown against a lovely desktop:

Ubuntu running, NAT-ed, no worries. Notice the CPU utilization is very low. Important, pay attention for later on when we discuss the emulation only mode. I'm nagging you, but please bear with my ultra-turbo education.

You can also scale the console window to fit guest; this is another option available in the viewer menu. Then, you can also use the preferences to autoscale the consoles. Here are Puppy and Ubuntu, side by side:

Here's the system usage;最小限。 With this kind of smart consumption, you can generously overcommit and enjoy more virtual machines than your physical hardware would normally permit, with a single machine bound to a core.

Eventually, you can also delete a machine and its associated files:

Now that we're comfortable with basic usage, let's do some command line.

コマンド ライン

Now, let's see the infamous command line. For example, using virsh to list all available virtual machines.

virsh 'list --all'

Here's the sequence of commands to get a virtual machine created and running using virt-install. Please note that I'm not a KVM guru. You have template code in the man pages, and you can find a billion examples online.

The actual command is:

virt-install --connect qemu://system -n puppy -r 512 -f puppy.img -c lupu-520.iso --vnc --noautoconsole --os-type linux --accelerate --network=network:default

では、ここには何がありますか？

--connect qemu:///system defines the type of hypervisor used. system is used when running on bare metal kernel as root. session is used for non-root users. You can recall the differences in VMM, with Usermode, right-click connect, you get the idea. This also works for xen.

-n puppy is a unique virtual machine name. You can use virsh to rename or delete the guest if you're not satisfied with your choice.

-r 512 specifies the RAM.

-f specifies a disk file. In this case, I'm using a puppy.img disk image I created using a dd command. To conserve space, I created the image as a sparse file. You can learn more about how to create sparse files in my second cool Linux hacks article.

-c specifies the CD-ROM, either a physical device or an ISO image.

--vnc setups a guest console and exports it as a VNC server. --noautoconnect will not automatically open the console when the virtual machine is powered on.

--os-type specifies the operating system for the guest.

--accelerate allows KVM to use optimization functions that accelerate the performance of the guest system, if possible.

--network defines the network type. We're using the default connection.

We could have used many other functions, like setting up the number of cores or bounding the machine to a CPU core, setup a fixed MAC address for NIC, use no storage, change the keyboard map, and more.

All of these are listed in the man pages. Not really intuitive, but becomes easier with use. In fact, the learning curve is not that high, despite the initial impression. Once you get the hang of virt-install, you can fiddle with other commands, start and stop machines with virsh, and Bob's your uncle.

その他

A few more details worth mentioning.

Running in pure emulation mode (don't)

I told you this is completely inefficient. Now, a proof. For instance, take a look at Host Details Overview. In the Performance graph, CPU usage is max. at 100% of the available resources for the virtual machine, in this case one core, which translates to 25% utilization. Almost like DOS. This means that four virtual machines, no matter how tiny or hungry, will cripple a four-core host.

Additionally, the performance will be horrible. If an Ubuntu guest takes approx. one minute to boot with the image loaded on the internal 7,200rpm disk when running with Intel-VT extensions enabled, it takes about 20 minutes to do that in the emulation mode. I think this is a downside of QEMU/KVM compared to other solutions, which still offer fairly decent performance even without CPU extensions.

More good reading

I highly recommend the KVM series on howtoforge.com, linked above.

Then, you should also read these:

Discover the Linux Kernel Virtual Machine

KVM management tools

結論

I sincerely hope you've liked this tutorial. It's fairly long and detailed, even though it documents only the first steps to get started with KVM. But it does offer an insight into a lot of things, including initial setup and prerequisites, command-line and GUI tools, management options, how to create and run multiple virtual machines, and more.

There are going to be a ton of sequels, which will cover advanced storage and networking configurations, including bridged networking, how to clone and migrate virtual machines, how to resolve conflicts with other virtualization software and more. Not strictly related, there are also going to be a whole bunch of articles on Xen and ESXi, too.

As you can see, KVM is quite powerful, even though it may appear a bit daunting to new users. Its management options are definitely less newb-friendly than some of its rival, but luckily, some of the fear and confusion have been spirited away today.

Well, I guess that would be all.楽しみ！

乾杯。