1. Computer Laboratory Virtualizing the Data Center with Xen Steve Hand University of Cambridge and XenSource

2. What is Xen Anyway? Open source hypervisor –run multiple OSes on one machine –dynamic sizing of virtual machine –much improved manageability Pioneered paravirtualization –modify OS kernel to run on Xen –(applications mods not required) –extremely low overhead (~1%) Massive development effort –first (open source) release 2003. –today have hundreds of talented community developers

3. Problem: Success of Scale-out OS+app per server provisioning leads to server sprawl Server utilization rates <10% Expensive to maintain, house, power, and cool Slow to provision, inflexible to change or scale Poor resilience to failures

4. Solution: Virtualize With Xen Consolidation: fewer servers slashes CapEx and OpEx Higher utilization: make the most of existing investments Instant on provisioning: any app on any server, any time Robustness to failures and auto-restart of VMs on failure

5. Further Virtualization Benefits Separating the OS from the hardware –Users no longer forced to upgrade OS to run on latest hardware Device support is part of the platform –Write one device driver rather than N –Better for system reliability/availability –Faster to get new hardware deployed Enables Virtual Appliances –Applications encapsulated with their OS –Easy configuration and management

6. Virtualization Possibilities Value-added functionality from outside OS: –Fire-walling / network IDS / inverse firewall –VPN tunnelling; LAN authentication –Virus, mal-ware and exploit scanning –OS patch-level status monitoring –Performance monitoring and instrumentation –Storage backup and snapshots –Local disk as just a cache for network storage –Carry your world on a USB stick –Multi-level secure systems

7. This talk Introduction Xen 3 –para-virtualization, I/O architecture –hardware virtualization (today + tomorrow) XenEnterprise: overview and performance Case Study: Live Migration Outlook

8. Xen 3.0 (5th Dec 2005) Secure isolation between VMs Resource control and QoS x86 32/PAE36/64 plus HVM; IA64, Power PV guest kernel needs to be ported –User-level apps and libraries run unmodified Execution performance close to native Broad (linux) hardware support Live Relocation of VMs between Xen nodes Latest stable release is 3.1 (May 2007)

9. Xen 3.x Architecture Event Channel Virtual MMUVirtual CPU Control IF Hardware (SMP, MMU, physical memory, Ethernet, SCSI/IDE) Native Device Drivers GuestOS (XenLinux) Device Manager & Control s/w VM0 GuestOS (XenLinux) Unmodified User Software VM1 Front-End Device Drivers GuestOS (XenLinux) Unmodified User Software VM2 Front-End Device Drivers Unmodified GuestOS (WinXP)) Unmodified User Software VM3 Safe HW IF Xen Virtual Machine Monitor Back-End HVM x86_32 x86_64 IA64 AGP ACPI PCI SMP Front-End Device Drivers

10. Para-Virtualization in Xen Xen extensions to x86 arch –Like x86, but Xen invoked for privileged ops –Avoids binary rewriting –Minimize number of privilege transitions into Xen –Modifications relatively simple and self-contained Modify kernel to understand virtualised env. –Wall-clock time vs. virtual processor time Desire both types of alarm timer –Expose real resource availability Enables OS to optimise its own behaviour

11. x86 CPU virtualization Xen runs in ring 0 (most privileged) Ring 1/2 for guest OS, 3 for user-space –GPF if guest attempts to use privileged instr Xen lives in top 64MB of linear addr space –Segmentation used to protect Xen as switching page tables too slow on standard x86 Hypercalls jump to Xen in ring 0 –Indirection via hypercall page allows flexibility Guest OS may install fast trap handler –Direct user-space to guest OS system calls

12. Para-Virtualizing the MMU Guest OSes allocate and manage own PTs –Hypercall to change PT base Xen must validate PT updates before use –Allows incremental updates, avoids revalidation Validation rules applied to each PTE: –1. Guest may only map pages it owns* –2. Pagetable pages may only be mapped RO Xen traps PTE updates and emulates, or unhooks PTE page for bulk updates

13. SMP Guest Kernels Xen 3.x supports multiple VCPUs –Virtual IPIs sent via Xen event channels –Currently up to 32 VCPUs supported Simple hotplug/unplug of VCPUs –From within VM or via control tools –Optimize one active VCPU case by binary patching spinlocks NB: Many applications exhibit poor SMP scalability – often better off running multiple instances each in their own OS

14. Performance: SPECJBB Average 0.75% overhead Native 3 GHz Xeon 1GB memory / guest 2.6.9.EL vs. 2.6.9.EL-xen 3 GHz Xeon 1GB memory / guest 2.6.9.EL vs. 2.6.9.EL-xen

15. Performance: dbench < 5% Overhead up to 8 way SMP

16. Kernel build 32b PAE; Parallel make, 4 processes per CPU 5% 8% 13% 18% 23% Source: XenSource, Inc: 10/06 # Virtual CPUs

17. I/O Architecture Xen IO-Spaces delegate guest OSes protected access to specified h/w devices –Virtual PCI configuration space –Virtual interrupts –(Need IOMMU for full DMA protection) Devices are virtualised and exported to other VMs via Device Channels –Safe asynchronous shared memory transport –Backend drivers export to frontend drivers –Net: use normal bridging, routing, iptables –Block: export any blk dev e.g. sda4,loop0,vg3 (Infiniband / Smart NICs for direct guest IO)

18. Isolated Driver Domains Event Channel Virtual MMUVirtual CPU Control IF Hardware (SMP, MMU, physical memory, Ethernet, SCSI/IDE) Native Device Driver GuestOS (XenLinux) Device Manager & Control s/w VM0 Native Device Driver GuestOS (XenLinux) VM1 Front-End Device Drivers GuestOS (XenLinux) Unmodified User Software VM2 Front-End Device Drivers GuestOS (XenBSD) Unmodified User Software VM3 Safe HW IF Xen Virtual Machine Monitor Back-End Driver Domain

19. Isolated Driver VMs Run device drivers in separate domains Detect failure e.g. –Illegal access –Timeout Kill domain, restart E.g. 275ms outage from failed Ethernet driver 0 50 100 150 200 250 300 350 05 10 15 20 25 30 35 40 time (s)

20. Hardware Virtualization (1) Paravirtualization… –has fundamental benefits… (c/f MS Viridian) –but is limited to OSes with PV kernels. Recently seen new CPUs from Intel, AMD –enable safe trapping of difficult instructions –provide additional privilege layers (rings) –currently shipping in most modern server, desktop and notebook systems Solves part of the problem, but…

21. Hardware Virtualization (2) CPU is only part of the system –also need to consider memory and I/O Memory: –OS wants contiguous physical memory, but Xen needs to share between many OSes –Need to dynamically translate between guest physical and real physical addresses –Use shadow page tables to mirror guest OS page tables (and implicit no paging mode) Xen 3.0 includes software shadow page tables; future x86 processors will include hardware support.

22. Hardware Virtualization (3) Finally we need to solve the I/O issue –non-PV OSes dont know about Xen –hence run standard PC ISA/PCI drivers Just emulate devices in software? –complex, fragile and non-performant… –… but ok as backstop mechanism. Better: –add PV (or enlightened) device drivers to OS –well-defined driver model makes this relatively easy –get PV performance benefits for I/O path

23. Xen 3: Hardware Virtual Machines Enable Guest OSes to be run without modification –E.g. legacy Linux, Solaris x86, Windows XP/2003 CPU provides vmexits for certain privileged instrs Shadow page tables used to virtualize MMU Xen provides simple platform emulation –BIOS, apic, iopaic, rtc, net (pcnet32), IDE emulation Install paravirtualized drivers after booting for high- performance IO Possibility for CPU and memory paravirtualization –Non-invasive hypervisor hints from OS

24. Native Device Drivers Control Panel (xm/xend) Front end Virtual Drivers Linux xen64 Device Models Guest BIOS Unmodified OS Domain N Linux xen64 Callback / Hypercall VMExit Virtual Platform 0D Backend Virtual driver Native Device Drivers Domain 0 Event channel 0P 1/3P 3P I/O: PIT, APIC, PIC, IOAPICProcessorMemory Control InterfaceHypercallsEvent ChannelScheduler Guest BIOS Unmodified OS VMExit Virtual Platform 3D HVM Architecture Guest VM (HVM) (32-bit mode) Guest VM (HVM) (64-bit mode)

25. Progressive Paravirtualization Hypercall API available to HVM guests Selectively add PV extensions to optimize –Network and Block IO –XenAPIC (event channels) –MMU operations multicast TLB flush PTE updates (faster than page fault) page sharing –Time (wall-clock and virtual time) –CPU and memory hotplug

26. Native Device Drivers Control Panel (xm/xend) Front end Virtual Drivers Linux xen64 Device Models Guest BIOS Unmodified OS Domain N Linux xen64 Callback / Hypercall VMExit Virtual Platform 0D Backend Virtual driver Native Device Drivers Domain 0 Event channel 0P 1/3P 3P I/O: PIT, APIC, PIC, IOAPICProcessorMemory Control InterfaceHypercallsEvent ChannelScheduler FE Virtual Drivers Guest BIOS Unmodified OS VMExit Virtual Platform FE Virtual Drivers 3D PIC/APIC/IOAPIC emulation Guest VM (HVM) (32-bit mode) Guest VM (HVM) (64-bit mode) Xen 3.0.3 Enhanced HVM I/O

27. HVM I/O Performance Measured with ttcp (1500 byte MTU) Emulated I/O PV on HVM Pure PV Source: XenSource, Sep 06

28. Native Device Drivers Control Panel (xm/xend) Front end Virtual Drivers Linux xen64 Guest BIOS Unmodified OS Domain N Linux xen64 Callback / Hypercall VMExit Virtual Platform 0D Guest VM (HVM) (32-bit mode) Backend Virtual driver Native Device Drivers Domain 0 Event channel 0P 1/3P 3P I/O: PIT, APIC, PIC, IOAPICProcessorMemory Control InterfaceHypercallsEvent ChannelScheduler FE Virtual Drivers Guest BIOS Unmodified OS VMExit Virtual Platform Guest VM (HVM) (64-bit mode) FE Virtual Drivers 3D IO Emulation Future plan: I/O stub domains. Device Models PIC/APIC/IOAPIC emulation

29. HW Virtualization Summary CPU virtualization available today –lets Xen support legacy/proprietary OSes Additional platform protection imminent –protect Xen from IO devices –full IOMMU extensions coming soon MMU virtualization also coming soon: –avoid the need for s/w shadow page tables –should improve performance and reduce complexity Device virtualization arriving from various folks: –networking already here (ethernet, infiniband) –[remote] storage in the works (NPIV, VSAN) –graphics and other devices sure to follow…

30. What is XenEnterprise? Multi-Operating System –Windows, Linux and Solaris Bundled Management Console Easy to use –Xen and guest installers and P2V tools –For standard servers and blades High Performance Paravirtualization –Exploits hardware virtualization –Per guest resource guarantees Extensible Architecture –Secure, tiny, low maintenance –Extensible by ecosystem partners Ten minutes to Xen

31. Get Started with XenEnterprise Ten minutes to Xen »Packaged and supported virtualization platform for x86 servers »Includes the Xen TM Hypervisor »High performance bare metal virtualization for both Windows and Linux guests »Easy to use and install »Management console included for standard management of Xen systems and guests »Microsoft supports Windows customers on XenEnterprise »Download free from www.xensource.com and try it out!www.xensource.com

32. Resource usage statisticsWindows 2003 ServerWindows XP Easy VM CreationClone and Export VMsRed Hat Linux 4 A few XE 3.2 Screenshots…

33. SPECjbb2005 Sun JVM installed

34. w2k3 SPECcpu2000 integer (multi)

35. w2k3 Passmark-CPU results vs native

36. w2k3 Passmark memory vs native

37. Performance: Conclusions Xen PV guests typically pay ~1% (1VCPU) or ~2% (2VCPU) compared to native –ESX 3.0.1 pays 12-21% for same benchmark XenEnterprise includes: –hardened and optimized OSS Xen –proprietary PV drivers for MS OSes –matches or beats ESX performance in nearly all cases Some room for further improvement…

38. Live Migration of VMs: Motivation VM migration enables: –High-availability Machine maintenance –Load balancing Statistical multiplexing gain Xen

39. Assumptions Networked storage –NAS: NFS, CIFS –SAN: Fibre Channel –iSCSI, network block dev –drdb network RAID Good connectivity –common L2 network –L3 re-routeing Xen Storage

40. Challenges VMs have lots of state in memory Some VMs have soft real-time requirements –E.g. web servers, databases, game servers –May be members of a cluster quorum Minimize down-time Performing relocation requires resources Bound and control resources used

41. Stage 0: pre-migration Stage 1: reservation Stage 2: iterative pre-copy Stage 3: stop-and-copy Stage 4: commitment Relocation Strategy VM active on host A Destination host selected (Block devices mirrored) Initialize container on target host Copy dirty pages in successive rounds Suspend VM on host A Redirect network traffic Synch remaining state Activate on host B VM state on host A released

42. Pre-Copy Migration: Round 1

47. Pre-Copy Migration: Round 2

52. Pre-Copy Migration: Final

53. Pages that are dirtied must be re-sent –Super hot pages e.g. process stacks; top of page free list –Buffer cache –Network receive / disk buffers Dirtying rate determines VM down-time –Shorter iterations less dirtying … Writeable Working Set

54. Rate Limited Relocation Dynamically adjust resources committed to performing page transfer –Dirty logging costs VM ~2-3% –CPU and network usage closely linked E.g. first copy iteration at 100Mb/s, then increase based on observed dirtying rate –Minimize impact of relocation on server while minimizing down-time

55. Web Server Migration

56. Iterative Progress: SPECWeb 52s

57. Iterative Progress: Quake3

58. Quake 3 Server relocation

59. Xen 3.1 Highlights Released May 18 th 2007 Lots of new features: –32-on-64 PV guest support run PAE PV guests on a 64-bit Xen! –Save/restore/migrate for HVM guests –Dynamic memory control for HVM guests –Blktap copy-on-write support (incl checkpoint) –XenAPI 1.0 support XML configuration files for virtual machines VM life-cycle management operations; and Secure on- or off-box XML-RPC with many bindings.

60. Xen 3.x Roadmap Continued improved of full-virtualization –HVM (VT/AMD-V) optimizations –DMA protection of Xen, dom0 –Optimizations for (software) shadow modes Client deployments: support for 3D graphics, etc Live/continuous checkpoint / rollback –disaster recovery for the masses Better NUMA (memory + scheduling) Smart I/O enhancements XenSE : Open Trusted Computing

61. Backed By All Major IT Vendors * Logos are registered trademarks of their owners

62. Summary Xen is re-shaping the IT industry Commoditize the hypervisor Key to volume adoption of virtualization Paravirtualization in the next release of all OSes XenSource Delivers Volume Virtualization XenEnterprise shipping now Closely aligned with our ecosystem to deliver full- featured, open and extensible solutions Partnered with all key OSVs to deliver an interoperable virtualized infrastructure

63. Thanks! EuroSys 2008 Venue: Glasgow, Scotland, Apr 2-4 2008 Important dates: –1–14 Sep 2007: submission of abstracts –2–21 Sep 2007: submission of papers –2–20 Dec 2007: notification to authors More details at http://www.dcs.gla.ac.uk/Conferences/EuroSys2008