1. Zen and the Art of Virtualization Paul Barham, et al. University of Cambridge, Microsoft Research Cambridge Published by ACM SOSP’03 Presented by Tina Swenson CS533 – March 5, 2007

2. Roadmap  Virtualization History  About Xen  Virtual Machine Interfaces  Design  Evaluation  Xen Today

3. Brief History of Virtualization  1960s IBM System/360 Model 67 Mainframe  Term Hypervisor born out of Supervisor.  Full Virtualization (Disco)  Unmodified OS/application code  Performance hit because hypervisor mediation. x86 architecture problems.  Paravirtualization (Xen)  OS cooperates with hypervisor.  OS code must be modified for this cooperation.

4. About Xen  A high-performance, resource-managed x86 virtual machine monitor (VMM).  100 guest OS instances running industry standard apps and services.  The hypervisor (VMM) sitting just above the hardware at a higher privilege mode than the OSes.

5. Virtual Machine Interface  Memory Management  Guest OSes are responsible for the hardware page tables.  Xen exists in the top section of every address space.  Avoids a TLB flush when entering and leaving the hypervisor.  Disco maintains shadow page tables to handle TLB misses and remain invisible.

6. Virtual Machine Interface  CPU  The OS is no longer at the most privileged level of the system.  Hypervisor in Ring 0. OS in Ring 1. App code in Ring 3.  Privileged instructions must be validated and executed within Xen.  Disco vs Xen  x86 exceptions map to Xen exception handling tables.

7. Virtual Machine Interface  Device IO  Device abstractions provided.  IO data moves between domains via shared memory and asynchronous buffer-descriptor rings.

8. Control Transfer  Hypercall  Synchronous trap by domain into hypervisor to perform a privileged op.  Events  Asynchronous events, replaces device interrupts.

9. IO Data Transfer  Circular queue of descriptors  Allocated by a domain, accessible by Xen.  Producer-Consumer

10. Subsystem Virtualization  CPU Scheduling  Domains scheduled via Borrowed Virtual Time.  Virtual Address Translation  Xen registers guest OS page tables with the MMU and allows the Guest OS read-only access. Xen performs page table updates.  No shadow page tables, as with Disco.  Physical Memory  Statically partitioned between domains.  Mapping physical to hardware addresses is up to Guest OS.  Balloon driver – Pages passed between Xen and XenoLinux.  Other  Time and Timers  Network  Disk

11. Performance  Linux L VMware Workstation V XenoLinux X User-Mode Linux U

12. Performance  Linux L VMware Workstation V XenoLinux X User-Mode Linux U

13. Performance  Linux L VMware Workstation V XenoLinux X User-Mode Linux U

14. Concurrent VM  Linux vs XenoLinux

15. More Evaluation  Performance Isolation  4 VMs. 2 normal, 2 nasty  Scalability  Can Xen run 100 domains?

16. Xen Today  Active Project  University of Cambridge  SourceForge  In Industry  IBM, HP, Intel, Red Hat, AMD and Novell  XenSource