Introduction to Virtualization Shruti

1
Introduction to Virtualization Shrut
i
2
Contents

• Introduction
• History of Virtual machines
• Benefits
• Example Xen
• Architecture
• Performance
• Screen shots
• Conclusion

3
History of Virtualization

• Introduced in 1960's to allow partitioning of
  large, mainframe h/w (scarce and expensive)
• Mini computers and pc's provided more efficient
  and affordable way to distribute processing
  power.
• By 1980's Virtualization was not employed
• In 1990's researchers saw how vm could solve some
  problems like proliferation of less expensive h/w.

4
Traditional Enterprise IT

• 6/7 million servers manufactured every year.
• Most of them are Intel x86 architecture
• Single app runs on OS across the data centers.
• Is too costly a h/w.

App OS
APP OS
5-10
5
Trend in Enterprise IT
App
App
App
Os
Os
Os
Virtualization
Hardware
Running applications side by side.
View of a virtualized m/c.
6
Definition

• Abstraction layer that decouples the physical
  hardware from the OS.
• Multiple virtual machines, heterogeneous OS's,
  run in isolation, side-by-side, same physical
  machine.
• Each VM has its own virtual hardware (e.g., RAM,
  CPU, NIC, etc.) on which OS and app's are loaded.

7
Benefits of Virtualization

• Partitioning
• Multiple apps and OS in 1 physical system.
• Resources allocated to VM in a controlled manner
• Isolation
• Isolated from host and other VM's.
• No data leak across VM's. App's communicate over
  n/w
• Encapsulation
• 1 VM env saved as 1 file, easy to move and copy.
• 1 standard virtual h/w to the app- guarantees
  compatibility.

8
Xen

• Is a VM monitor for X86.
• Supports execution of multiple guest Operating
  sys.
• Is an Open Source software.
• Each OS performs its own paging using its memory
  reservation and disk allocation.
• Challenges
• Isolation
• Performance overhead made small
• Supporting multiple OS.

9
Virtual Machine Interface
X86 Architecture
Memory Management
CPU
Device I/O
10
Memory Management

• Most difficult part Hypervisor, porting each
  guest OS
• Hypervisor
• Thin layer of software running on the hardware.
• Supports creation of partitions
• Each partition is a virtual machine
• Each partition has one or more virtual processors
• Partitions can own or share hardware resources.
• Software running in partition is called a guest.
• Enforces memory access rules.
• Enforces policy for CPU usage
• Virtual processors are scheduled on real
  processors.
• Enforces ownership of other devices.
• Provides simple inter-partition messaging
• Messages appear as interrupts

11
MM cont..

• Each guest OS performs its own paging using its
  own memory reservation and disk allocation.
• Done using TLB. TLB misses processed by walking
  page table structure in hardware.
• So, all valid page translations for current
  address space should be on the h/w accessible
  page.
• Each time guest OS requires a new page table(new
  process) it allocates a page from memory.
• Updates must be made and validated by Xen.
• Xen occupies top 64MB region of each addr space,
  not accessed by guest OS.

12
CPU

• CPU virtualization has several implications on
  Os.
• Due to the Hypervisor below the OS, OS is not
  given the top level privilege.
• So, OS should be modified to run at lower
  privilege
• X86 supports 4 privilege levels called
  rings.(0-3)
• OS code executes on ring 0 and application code
  on ring 3. Ring 1 and 2 are not used. All such OS
  can be ported to Xen easily.
• So, OS does not execute privileged instructions
  and remains isolated from running applications.

13
CPU cont..

• When OS tries to execute privileged instruction,
  its failed by processor.
• Exceptions, memory faults and s/w traps are
  virtualized using a table.
• Table handler for each exception registered
  with Xen.
• 2 frequent exceptions system call, page faults.
• System call accessed by processor.
• Page faults delivered via Xen.

14
Device I/O

• Xen exposes simple device abstractions.
• So, easy to design an efficient interface which
  is protected and isolated.
• I/O data transferred between domains using shared
  memory.
• High performance communication mechanism for
  passing buffer information through the system.

15
Cost of Porting OS to Xen

• XP required more modifications to its
  architecture.
• It uses many structures and unions for accessing
  PTE's.
• Each PTE has to be separately modified.
• Linux needed less modifications.
• It uses pre-processor macros to access PTE's.

16
Xen hypervisor with many guest os
17
Detail Design

• 2 Interaction mechanisms b/n Xen and Domain
• Hypercall b/n Domain to Xen.(synchronous)
• Events b/n Xen and Domain.(asynchronous)
• Data Transfer I/O rings.
• Ring is a circular queue of descriptors.
• Is allocated by domain but accessible by Xen.
• Descriptors contain the reference to data.
• Each ring accessed using 2 pairs of producer
  consumer pointers.
• Domains place request on ring, request pointer
  advances, Xen removes the request and handles
  them, advances consumer pointer.

18
Structure of I/O rings.
Request Producer Shared pointer updated by guest
Os
Request Consumer Private pointer in Xen
Response producer shared pointer updated by Xen.
Response Consumer Private pointer in guest OS
19
Subsystem Virtualization

• CPU Scheduling BVT scheduling algorithm.
• Virtual Address Translations
• Xen virtualizes memory with little overhead.
• Xen updates page tables to prevent guest OSes to
  make unacceptable changes.
• Xen registers guest OS page tables with MMU and
  restrict guest OSes to read-only access.
• Page table updates are passed to Xen via
  hypercall.

20
Subsystem Virtualization cont...

• Physical Memory
• Initial memory allocation for each domain
  specified at the time of creation.
• Memory is statically partitioned isolation.
• Max allowable reservation may be specified.
• Additional memory pages can be claimed from Xen
  if memory pressure within Xen increases.
• If domain wishes to save resources, reduce memory
  reservation by releasing memory pages back to Xen.

21
Xen-Screen shot
22
Conclusion

• Above were the advantages of virtualization with
  its working.
• We can now allow 100 Os to run on a single server
  with low cost.
• Xen is good for network-centric services.
• It is observed that the performance of Xen is as
  good as a single Linux system.

23
Questions