Virtualization in Cloud Computing

Virtualization in Cloud Computing

1. Virtualization

Overview
  • Operating System concept: The abstraction of available resources
  • Virtualization technologies encompass a variety of mechanisms and techniques used to address computer system problems such as security, performance, and reliability by decoupling the architecture and user-perceived behavior of hardware and software resources from their physical implementation. (https:/www.computer.org/csdl/mags/co/2005/05/r5028.html/)
Overview of virtualization
Virtualization requirements
  • Formal requirements for virtualizeable third generation architectures (Paper on D2L)
  • A virtual machine is taken to be an efficient, isolated duplicate of the real machine.
  • These notions can be explained through the idea of a virtual machine monitor.
  • Essential characteristics of VMM:
    • Essentially identical to the physical resource
    • Efficiency
    • Complete control of system resources (with regard to the processes running inside the VM)
Overview of virtualization
Virtual Machine Monitor
  • Virtualization Layer: The Virtual Machine Monitor (or its modern name: Hypervisor) provides an interface between hardware and virtual operating systems.
  • Type of hypervisors:
    • Bare-metal
    • Hosted
Types of virtualizations
Issues that virtualization can help with
  • Under-utilized resources
  • Complicated system management
  • Limited access to shared resources
  • Inefficient power consumption
  • Tight coupling with underlying resources
Virtualization versus multitasking versus multicore/hyperthreading
Virtualization versus multitasking versus multicore/hyperthreading

2. Types of virtualization

Overview
  • Platform Virtualization
  • Memory Virtualization
  • Desktop Virtualization
  • Application Virtualization
  • Network Virtualization
  • Storage Virtualization
Platform virtualization
  • Full Virtualization
  • Para Virtualization
  • Hardware assisted virtualization
  • OS level virtualization
Without virtualization
  • x86 offers four levels of privilege (Ring 0 through 3)
  • OS needs to have access to hardware and run on ring 0
  • Application runs on ring 3, gain access to hardware by trapping into kernel mode for privileged instructions.
  • Virtualizing x86 requires a layer under OS (which already at lowest level) to create and manage the VM
  • Sensitive instructions must be executed in ring 0
Levels of privilege in x86 architecture
Full virtualization
  • Guess OS is unaware of host OS.
    • VMM provides virtual BIOS, virtual devices, and virtual memory management.
  • Non-critical instructions run directly on hardware.
  • Runtime translation of critical non-virtualizable instructions happens in the hypervisor.
  • Provide best isolation and security at the cost of performance.
x86's levels of privilege for full virtualization
Para virtualization
  • Thin layer interfaces between each guest OS and underlying hardware.
  • Need guest kernel modification.
  • No need of runtime translation for critical instructions.
  • Superior in performance.
  • Requires expertise to patch the kernels.
x86's levels of privilege for para virtualization
Hardware-assisted virtualization
  • Hardware provides support to run instructions independently.
    • Intel Virtualization Technology (VT-x)
    • AMD Virtualization Technology (AMD-V)
  • No need to patch the kernels.
  • Runtime translation not required.
  • Better performance in comparison to other variants.
  • Greater stability
x86's levels of privilege for hardware-assisted virtualization
Virtualization at OS level
  • Same OS for both host and guest machines.
  • User space is completely isolated.
  • High performance.
  • Extremely light-weight.
Memory virtualization
  • How to share physical system memory and dynamically allocating it to virtual machines.
  • Guess OS maps virtual memory space (of VM) to physical memory space (of VM).
  • VMM translates physical memory space (of VM) to physical memory space (of main machine), but also enables direct mapping (shadow table) to avoid overhead.
Memory virtualization
Summary
  • Comparative issues
    • Guess modification/compatibility
    • Performance
    • Guest OS hypervisor/independent

=== “Full/binary translation”



1
2
3

 
- Unmodified guest OS, excellent compatibility
- Good
- Yes

=== “Hardware assisted”



1
2
3

 
- Unmodified guess OS, excellent compatibility
- Fair
- Yes

=== “OS assisted/para”



1
2
3
4
5
6

 
- Guest OS codified to run Hypercall
    - Cannot run off native hardware or other hypervisors
    - Poor compatibility
- Better on certain cases
- Xen Linux runs only Xen Hypervisor
    - VM-Linux is hypervisor-agnostic
Desktop and application virtualization
  • Desktop and Applications run on servers.
  • Stateless thin clients connected to servers.
  • Efficient system management.
  • Requires high-end servers for system stability
Network and storage virtualization
  • Similar idea of providing an abstraction layer to the physical infrastructures
  • In networks, abstraction will
    • Be at the level of routers, switches, gateway, firewalls, load balancers, …
    • Enabled by software-defined networking
  • In storage, single storage backends can be used for different requirements
    • Ephemeral
    • Persistent
    • Specialize storage backends

3. Virtualization: concept of overcommits

Overview
  • Allocating more than the available physical resources to the Guest OS
  • Common types of overcommit:
    • CPU
    • Memory
    • Storage
Virtualization: concept of overcommits
  • Advantages:
    • Favorable economic model
    • Efficient resource utilization
    • Support green computing
  • Disadvantages:
    • Performance loss or unstable system response
    • Complex system understanding
    • VM shutdown by the hypervisor
Virtualization: CPU and memory overcommits
  • Allows more virtual CPUs than physically available
    • Openstack KVM: overcommit-number = 16.0
  • Allow more memory than physically available
    • Openstack KVM: overcommit-number = 1.5GB
Virtualization hypervisors
  • Contribution from industry and academia
  • Xen: Project from Cambridge Computer Laboratory
  • VMware: Commercial product
    • Also comes from academic research (see Mendel Rosenblum ACM)
  • KVM: Initiated by the Open Virtualization Alliance, later dissolved and is now managed by the Linux Foundation
  • Qemu: Open source machine emulator and virtualizer

4. Virtualization in the cloud: OpenStack

Logical architecture
  • OpenStack Design
  • Module types:
    • Daemon: runs as a background process.
    • Script: installs a virtual environment and runs tests.
    • Command-line interface (CLI): enables users to submit API calls to OpenStack services
Openstack Logical Architecture
  • Compute: Provision compute instances
  • Keystone: Identity and authentication
  • Glance: Compute image repository
  • Neutron: Provision virtual/physical networks
  • Placement: Tracking inventory of resources available in a cloud
  • Cinder: Provision block storages
OpenStack: Compute
  • Message-based, “share-nothing” architecture
    • User-facing: REST API
    • Internal communication: RPC
  • Supported/controlled hypervisors:
    • Baremetal
    • Hyper-V
    • KVM
    • Linux Containers (LXC)
    • Quick Emulator (QEMU)
    • Virtuozzo
    • VMWare vSphere
    • zVM
Compute Nova: Architecture
Openstack: Cinder
  • Providing volumes to VMs, baremetal hosts, containers, etc.
Compute Nova: Architecture
Openstack: Neutron
  • Management: internal comm between OpenStack components, reachable only within the data center.
  • Guest: Used for VM data communication within the Cloud Deployment.
  • External: Provide VM with Internet access.
  • API: Exposed all the Stack’s API to the public.
Compute Nova: Architecture