The Open Source Xen Hypervisor

Introduction to the Open Source Xen Hypervisor Todd Deshane and Patrick Wilbur (Clarkson University) Stephen Spector (Citrix) June 22, 2008

Overview of the Day and Schedule

Unit 1: Virtualization and Xen Overview Morning Break Unit 2: Installing, Configuring, and Basic Usage Lunch Unit 3: Devices and Advanced Configuration Afternoon Break Unit 4: Security, Admin Tools, and Performance

Unit 1: Virtualization and Xen Overview

Unit 1
3

Unit 1: Virtualization and Xen Overview

Why Virtualize? Potential Pitfalls Types of Virtualization Xen Background Hardware-Assisted Virtualization Overview Xen Architecture Xen LiveCD Demonstration

Why Virtualize?

Consolidation of servers Support heterogeneous and legacy OSes Rapid deployment and provisioning Advantages

Testing/debugging before going into production Recovery and backup Load balancing

Potential Pitfalls

Hardware/software licensing complexities Added complexity per server

More to lose with a single server failure

Lack of IT staff with virtualization skills Security challenges Loss of deterministic hardware performance

Mitigated with configuration

Types of Virtualization

No Virtualization

Operating system running on native hardware Fully-emulate a hardware architecture

Emulation

Unmodified guest OSes Examples: QEMU, Bochs

Can be different than actual hardware architecture

Full

Simulate the base hardware architecture Unmodified guest OSes Examples: VMware, VirtualBox, QEMU + kqemu, MS Virtual PC, Parallels, Xen/KVM (hardware-assisted)
7

Binary re-writing Hardware-assisted virtualization

Types of Virtualization

Para

Abstracted base architecture Modified guest OSes Examples: Xen, UML, Lguest Shared kernel (and architecture), separate user spaces Homogeneous guest OSes Examples: OpenVZ, Linux-VServer, Solaris Containers, FreeBSD Jails

Operating System Level

Xen Background

Paravirtualization (PV)

High performance (claim to fame) With hardware support Hardware Virtual Machine (HVM) Open source Standalone hypervisor
Citrix Xen Server Virtual Iron Solaris xVM Oracle VM

Full Virtualization

Advantages

Hardware-Assisted Virtualization

IBM mainframe to commodity x86

Co-evolution of hardware and software AMD-V, Nested Page Tables (NPT) VT-x, VT-d, Extended Page Tables (EPT) AMD-V, VT-x Input/Output Memory Management Unit (IOMMU) NPT, EPT Smart devices

x86 AMD/Intel

Processor

Memory

I/O

Virtualization-aware NICs, graphics cards, etc.

Xen Architecture

Domain0 Management VM

Guest1

Guest2

GuestN

Xen Hypervisor

Physical Hardware
11

Xen Hypervisor Role

Thin, privileged abstraction layer Provides generic classes of devices

CPU, memory, disk, network, etc.

Scheduling

12

Domain0 Role

Creates and manages guest VMs Interacts with the Xen hypervisor

Xend (Xen daemon) xm commands Xenstore xenconsoled

Abstraction for guest ttys

Supplies device and I/O services

Runs backend drivers Provides guest storage

13

Xenstore Overview

Database of configuration information Used by Domain0 to access guest state Guest domain drivers can write to xenstore Reference:

http://wiki.xensource.com/xenwiki/XenStoreReference

14

Xen LiveCD Demo

xm list xm create xm shutdown xm reboot xm destroy xm pause xm unpause xm console

15

Unit 1: Virtualization and Xen Overview

Questions?
16

Morning Break

Break Time
17

Unit 2: Installing, Configuring, Basic Use

Unit 2
18

Unit 2: Installing, Configuring, Basic Use

Hands-on Installation Demos Guest Installation Distro-specific Guest Installation Tools Interacting with Guests

19

Hands-on Installation Demos

Build from source

Mercurial overview

Peer to peer version control hg serve, hg pull (-u), hg update, hg commit, etc. Supports tags, hg fetch extension in .hgrc As a user, hg clone is usually enough Example:
-

Source tarball of stable releases also available

hg clone http://xenbits.xensource.com/xen-unstable.hg hg clone http://xenbits.xensource.com/linux-2.6.18-xen.hg cd xen-unstable.hg make world make install

Distribution-supported packages preferred

Integrates better with distribution Security and bug fixes from distribution maintainers

20

Open Source Distributions

CentOS/Fedora Note: No Domain0 kernel in Fedora 9

21

During Installation: Customize Now

22

Add Virtualization Group

23

Optional Virtualization Packages

24

Open Source Distributions

OpenSUSE

25

During Installation: Software Selection

26

Xen Software Packages

27

Open Source Distributions

Debian/Ubuntu

Install with apt or synaptic Meta-packages

Ubuntu
-

ubuntu-xen-server ubuntu-xen-desktop

Gentoo

Install with portage

May need to unmask the Xen packages

OpenSolaris

xVM packages available Xen 3.X package support as of NetBSD 4.0

NetBSD

28

Commercial Distributions

Red Hat Enterprise Linux (RHEL) SUSE Linux Enterprise Server (SLES) Virtual Iron Oracle VM Xen Server

Express Edition (demo) Standard Edition Enterprise Edition

29

Guest Installation

Guest Configuration Files PV Guest Config Example

PV Guest Example (demo) HVM Guest Example (demo)

HVM Guest Config Example

Pre-built Guest Images Converting VMware Images Distribution-specific Guest Installation Tools Interacting with Guests
30

Guest Configuration Files

Disk I/O and memory configuration parameters

disk

Array of 3-tuples
-

SCSI (sd) and IDE (hd) examples:

Device to export Device as seen by guest Access permission (r, w)

Xen Virtual Block Device (xvd) examples:

disk = ['phy:sda, sda, w', 'phy:/dev/cdrom, cdrom:hdc, r'] disk = ['tap:aio:hdb1, hdb1, w', 'phy:/dev/LV/disk1, sda1, w'] disk = ['phy:sda, xvda, w', 'phy:/dev/cdrom, cdrom:hdc, r'] disk = ['tap:aio:hdb1, xvdb1, w', 'phy:/dev/LV/disk1, xvda1, w']

memory

Integer in megabytes (MB) Example:

memory = 512

31

Guest Configuration Files

Network configuration parameters

vif

Array of virtual interface specifications Zero or more name=value entries for each interface Single interface examples

Multiple interface examples

vif=[' '] vif=['mac=00:16:3e:51:c2:b1'] vif=['mac=00:16:3e:36:a1:e9, bridge=eth0'] vif=[' ', ' '] vif= ['mac=00:16:3e:36:a1:e9, bridge=eth1', 'bridge=eth0']

dhcp, gateway, hostname, netmask

32

Guest Configuration Files

PV guest kernel configuration parameters

kernel

ramdisk

Use Domain0 kernel (external to guest) Use Domain0 kernel (external to guest) pygrub
-

bootloader

root

Use Xen-compatible kernel and initrd (internal to guest) Root file system device Append to kernel command line
-

Security issues

extra

extra=4 sets runlevel to 4

33

Guest Configuration Files

HVM guest configuration parameters

kernel

builder

Use hvmloader Use hvm

device_model boot

Usually QEMU device model (qemu-dm) Boot order

sdl

vnc

Simple DirectMedia Layer Built-in to Xen virtual frame buffer Virtual Network Computing

34

Guest Configuration Files

Deprecated parameters

cdrom
-

file:/ in disk=['file:/...']
-

Replaced by disk=['....cdrom:hdX...'] Replaced with disk=['tap:aio:/...'] Better to simply use the vif array

nics
-

35

PV Guest Config Example

kernel="/boot/vmlinuz-2.6.24-16-xen" ramdisk="/boot/initrd.img-2.6.24-16-xen" disk=['tap:aio:/xen/para.partition,xvda1,w'] memory=512 vif=[' '] root="/dev/xvda1"

36

PV Guest Example

Demo

37

HVM Guest Config Example

HVM guest example kernel="/usr/lib64/xen/boot/hvmloader" builder="hvm" device_model = "/usr/lib64/xen/bin/qemu-dm" disk=['phy:/xen/images/hvm.disk,hda,w', 'phy:/dev/cdrom,hdc:cdrom,r'] sdl=1 boot="dc" memory=512 vif=['type=ioemu,bridge=eth0']
38

HVM Guest Example

Demo

39

Pre-built Guest Images

rpath.com jailtime.org virtualappliances.net jumpbox.com

40

Converting VMware Images

qemu-img convert (demo)

41

Distribution-specific Tools

CentOS/Fedora

Virtual Machine Manager (virt-manager) and rpmstrap Yast Virtual Machine Management debootstrap and xen-tools quickpkg and domi

OpenSUSE

Debian/Ubuntu

Gentoo

42

Distribution-specific Tools

Virtual Machine Manager example

43

Main Window Create New

44

Name the Guest VM

45

PV Installation Media Choices

46

HVM Installation Media Choices

47

Choose Disk Storage

48

Memory and CPU Settings

49

Begin the Installation

50

Distribution-specific Tools

Yast Virtual Machine Management example

51

Main Window Add Guest VM

52

Guest Installation Choices

53

Start Installation

54

Interacting with Guests

SSH VNC FreeNX Xen console Xen virtual frame buffer (sdl) Remote Desktop (demo)

55

Unit 2: Installing, Configuring, Basic Use

Questions?
56

Lunch

Food Time
57

Unit 3: Devices and Advanced Config

Unit 3
58

Unit 3: Devices and Advanced Config

Quick LVM Refresher Network Storage Guest Image Files from Scratch Guest Save, Restore, and Live Migration Xen Device Models Network Configurations

59

Quick LVM Refresher

Logical Volume Manager (LVM) Provides abstraction above block devices Allows the system administrator to:
Span logical volumes across physical volumes Grow and shrink logical volumes Use various RAID levels Use copy-on-write

Why LVM?

Flexible

Feature-rich

Resize partitions (guest images) Density of logical volumes (guests) Copy-on-write snapshots
-

Backup and recovery

Quick LVM Refresher

Key terms

physical volume (PV) volume group (VG) logical volume (LV)

61

Quick LVM Refresher

LVM commands

pvcreate initializes a physical volume for LVM use

# pvcreate /dev/sda3 /dev/sda4 /dev/hda1 # vgcreate xen_vg /dev/sda3 /dev/sda4 /dev/hda1 # lvcreate -L 4G -n guest_partition xen_vg # lvextend -L 5G /dev/xen_vg/guest_partition # lvreduce -L 3G /dev/xen_vg/guest_partition

vgcreate creates a new volume group

lvcreate creates a new logical volume

lvextend grows a logical volume

lvreduce shrinks a logical volume

62

Quick LVM Refresher

LVM caveats

When extending a logical volume

When reducing a logical volume

Resize the underlying file system after the lvextend Resize the underlying file system before the lvreduce

Underlying file system support

Resize larger or smaller Resize online (while the file system is mounted)

File system-specific commands

ext2/3

resize2fs e2fsck tune2fs

63

Network Storage

ATA over Ethernet (AoE)

Exports block devices over the network Lightweight Ethernet layer protocol No built-in security Exports block devices over the network Scales with network bandwidth Network layer protocol Client and user-level security Exports file system over the network Well-known and widely used Network layer protocol Known performance issues as root file system
64

Internet Small Computer System Interface (iSCSI)

Network File System (NFS)

Network Storage

Network Block Device (NBD)

Exports block device over the network Scales with network bandwidth Network layer protocol Not recommended as root file system Advantages of block devices and file servers More difficult to setup and configure Examples:
Global Network Block Device (GNBD) Distributed Replicated Block Device (DRBD)

Cluster file systems

65

Guest Image Files from Scratch

Compressed tar image files

Smallest guest image file Best for backup and sharing More difficult to setup guest

Need existing place to extract to

Disk image files

Single file containing root and swap partitions Largest guest image file More commands needed Separate files for root and swap partitions Easiest to work with (standard commands) Slightly more image files to work with (separate swap)
66

Partition image files

Compressed Tar Image Files

Creation

tar -czpf /linux-root.tgz \ --exclude /proc \ --exclude /linux-root.tgz \ / Setup a partition, local volume, disk or partition image Extract with tar command Customize partition setup Configure network-specific details
/etc/hosts and /etc/hostname IP address setup

Using

/etc/fstab Kernel command line

67

Disk Image Files

Creation

Use dd to make a sparse or pre-allocated file Partition the image (fdisk) Make partitions available to the system (kpartx) Format the file systems (mkfs/mkswap) Populate root file system If no customization is needed, use directly If customization is needed

Using

Associate disk image with block device (losetup) Make partitions available to system (kpartx) Mount partitions in /dev/mapper (mount) Do customizations Unmount (umount), release partitions (kpartx -d), un-associate with block device (losetup -d)
68

Partition Image Files

Creation

Use dd to make a sparse or pre-allocated file Format each file system (mkfs/mkswap) Populate file system (if root partition) If no customization is needed, use directly If customization is needed
Mount with loop option (mount -o loop) Do customizations Unmount (umount)

Using

69

Guest Save, Restore, and Live Migration

Guest save and restore

Similar to hibernate and resume of physical PC Built-in Xen functionality and xm commands Cold static relocation Warm static (regular) migration Live migration

Guest relocation and migration

70

Guest Save and Restore

Save and restore commands

xm save

xm restore

Pauses and suspends (hibernates) a guest Saves guest state to a file on disk Restores a guest's state from disk Resumes execution of guest

71

Guest Relocation

Cold static relocation

Image and config files need to be manually copied from source to target Domain0 Hardware maintenance with less downtime Backup of guest images Shared storage not required More manual process Guests should be shut down during copy

Benefits

Limitations

72

Guest Migration

Warm static (regular) migration

xm migrate

Network connections to and from guest are interrupted and probably will timeout xm migrate --live
Copies a guest's state to a new Domain0 Repeatedly copies dirtied memory until transfer is complete Re-routes network connections

Pauses a guest Transfers guest state across network to a new Domain0 Resumes guest on destination host

Live migration

Network connections to and from guest remain active and uninterrupted

73

Guest Migration

Benefits

Load balancing Hardware maintenance with little or no downtime Relocation for various reasons Shared storage required Guests on same layer 2 network Sufficient resources needed on target machine CPU architectures need to match Some constraints on hypervisor version

Limitations

74

Xen Device Models

Split driver model QEMU device model No virtualization device support

PCI passthrough demo

75

Split Driver Model

Generic backends

Loaded in DriverDomain (often Domain0) Loaded in guest domain Connects to corresponding backend driver Guests use standard Xen virtual device drivers

Generic frontends (virtual devices)

Real device-specific drivers are in DriverDomain Devices are multiplexed to the guests

76

Split Driver Model

77

QEMU Device Model

Provides emulation of devices Provides exclusive access illusion to the guests Used primarily for HVM guests

device_model set to qemu-dm binary in guest config

78

No Virtualization Device Support

Guests granted full access to specific PCI devices The actual device driver runs in the guest Benefits

Highest performance for a device Useful when virtualization doesn't support a device System stability with buggy driver Not (yet) well-tested DriverDomain as backend can be tricky HVM guest support still limited Security considerations

Limitations

Without an IOMMU, guests can (direct memory access) DMA into main memory

79

PCI Passthrough

Demo

80

Network Configurations

Network Bridge Configuration Network Route Configuration Network NAT Configuration Host-only Networking Handling Multiple Interfaces Virtual Private Network (VPN)

Xen's vnet configuration

81

Network Bridge Configuration

A bridge relays traffic based on MAC address Behavior of guests in bridge mode

Appear transparently on the DriverDomain's network Access the network directly (through software bridge) Obtain IP address on the local Ethernet Set network-bridge and vif-bridge in xend config Bridge is default guest configuration

Configuration details

Can set bridge option manually

Secure with ebtables

MAC layer filtering (similar to iptables) brctl show

Troubleshoot with brctl (bridge-utils package)

Network Bridge Configuration

83

Network Route Configuration

A router relays traffic based on IP address Behavior of guests in route mode

Get routed to the Ethernet through the DriverDomain Access the network via DriverDomain Obtain IP address from DriverDomain

Manually or with DHCP relay

Configuration details

Set network-route and vif-route in xend config Set IP, netmask, and gateway in guest config ip_forwarding in DriverDomain Network traces and iptables configuration
84

Secure with iptables

Troubleshoot at IP layer

Network Route Configuration

85

Network NAT Configuration

Get NATed through the DriverDomain Access outgoing traffic transparently through NAT Are not visible from the outside (behind NAT router) Obtain internal IP from software NAT router Set network-nat and vif-nat in xend config Set IP, netmask, and gateway in guest config Uses MASQUERADE chain

Configuration details

Secured with iptables

Troubleshoot with internal and external traces

Network NAT Configuration

87

Host-only Networking

Supports local-only network traffic

Share data on private network Guests to other guests only Host (DriverDomain) to guests only Set up dummy bridge in DriverDomain Set bridge to dummy bridge in guest config

Dummy bridge connection options

Configuration details

Secure with ebtables (if needed) Troubleshoot with brctl and iptables
88

Host-only Networking (Guest to Guests)

89

Host-only Networking (Host to Guests)

90

Handling Multiple Interfaces

Configure the purpose of NICs in xend config Specify bridge device in guest config Create custom scripts that call network-* scripts

Set virtual device number (vifnum) Set network device (netdev) Set bridge device (bridge)

Add custom network script in xend config Add vif entries and specify bridge in guest config

91

Virtual Private Network (VPN)

Provides confidential communication on public network

Illusion of VPN client on VPN server's local network VPN implementation based on bridged network Installation methods Not well-tested (yet)
Kernel module Userspace daemon in DriverDomain

Xen's vnet configuration

92

Unit 3: Devices and Advanced Config

Questions?
93

Afternoon Break

Break Time
94

Unit 4: Security, Admin Tools, and Perf

Unit 4
95

Unit 4: Security, Admin Tools, and Perf

Security Basics Advanced Security: sHype and XSM Management APIs and Utilities Resources and Performance Future Directions

96

Security Basics

Standard system security practices apply Secure Domain0 and Xen Secure guest domains normally

97

Security Basics

Secure Xen Hypervisor

Secure Hypervisor (sHype) Xen Security Modules (XSM) Minimize software packages Minimize running services and open ports Use firewall and intrusion detection systems Similar to Domain0 software, services, and ports IOMMU for direct PCI device pass-through

Secure Domain0

Secure guests

Prevent insecure memory access through driver DMA

98

Advanced Security: sHype and XSM

Secure Hypervisor (sHype)

IBM Research project Role-Based Access Control (RBAC)

Mandatory Access Control (MAC)

Rights-based on privilege group

Virtual Trusted Platform Module (vTPM)

Labeled objects Privilege groups given access to specific object labels Enables remote attestation by digitally signing cryptographic hashes of software components Attestation means to affirm that some software or hardware is genuine or correct

Xen Security Modules (XSM)

National Security Agency (NSA) security framework Based on SELinux

Advanced Security: sHype

Demo

100

Management APIs and Utilities

APIs

libvirt Xen Application Programming Interface (API) Xen Common Information Model (CIM) virt-manager virsh XenMan (ConVirt) Enomalism Citrix XenServer Virtual Iron IBM Director (IBM Virtualization Manager extension)

Management utilities

101

Management APIs and Utilities

Demo

102

Resources and Performance

Gauging performance

xm top (demo) xentop xm mem-set xm mem-max xm vcpu-list xm vcpu-set xm vcpu-pin

Memory management

Virtual CPU management

103

Resources and Performance

Xen scheduler

Credit scheduler

Weight and cap

I/O schedulers

Noop scheduler Deadline scheduler Anticipatory scheduler (as) Complete fair queuing scheduler (cfq)

104

Resources and Performance

Macro benchmarks

Disk I/O

Network CPU

iozone, iometer, bonnie++ netperf, iperf, ttcp SPECint, kernbench

Web and database servers

SPECweb, Hammerhead, Apache bench (ab) Database benchmarks

Micro benchmarks and profiling

xenoprof, xenperf, xentrace, xenmon virtbench

Performance isolation suite (from Clarkson)

Future Directions

Smart devices

Virtualization-aware NICs, video cards AMD NPT, Intel EPT, etc.

Next generation hardware support

IOMMUs Trusted Platform Module (TPM) support Xen Domain0 inclusion in mainline Linux PV drivers for HVM guests

PVGPL Citrix Novell Various others

Microsoft enlightenments support

Further Information

Useful resources

Xen Community XenWiki Xen Mailing Lists Xen Bugzilla Xen Summit Xen source code Academic papers and conferences The Definitive Guide to the Xen Hypervisor (book) Running Xen: A Hands-On Guide to the Art of Virtualization (book)

107

Unit 4: Security, Admin Tools, and Perf

Questions?
108

Acknowledgments

Running Xen book co-authors

Jeanna Matthews Eli M. Dow Wenjin Hu Jeremy Bongio Brendan Johnson

Padmashree Apparao (Intel Research) Mark Williamson (University of Cambridge)

109

References

http://runningxen.com http://runningxen.com/mailman/listinfo/readers_runningxen.com http://www.usenix.org/publications/login/2007-02/pdfs/hand.pdf http://wiki.xensource.com/xenwiki/XenStoreReference http://portal.acm.org/citation.cfm?id=1281700.1281706&coll=&dl=ACM http://www.usenix.org/publications/login/2007-02/pdfs/griffin.pdf http://passat.crhc.uiuc.edu/dasCMP/papers/dasCMP07/paper01.pdf http://ieeexplore.ieee.org/iel5/4299339/4299340/04299347.pdf? isnumber=4299340&prod=CNF&arnumber=4299347&arSt=2&ared=2&arAuthor =Apparao%2C+Padma%3B+Makineni%2C+Srihari%3B+Newell%2C+Don http://workspace.globus.org/vtdc06/VTDC_files/programdraft.htm http://xen.org/files/xensummit_fall07/28_PadmaApparao.pdf http://opensolaris.org/os/project/libmicro/ http://www.computerworld.com/action/article.do? command=viewArticleBasic&taxonomyName=Storage&articleId=9081798&taxon omyId=19 http://weblog.infoworld.com/daily/archives/2008/04/top_3_gotchas_o.html http://news.zdnet.com/2100-3513_22-6191965.html http://www.cl.cam.ac.uk/research/srg/netos/xen/readmes/hg-cheatsheet.txt http://www.cuddletech.com/blog/pivot/entry.php?id=469
110

The Open Source Xen Hypervisor

Thank you for coming. Questions/Comments?