Virtual Machines vs. Containers Revisited - Part 2

Mobycast

Episode • Oct 16, 2019 • 49m

Sponsors

Show Details

In this episode, we cover the following topics:

Hypervisor implementations
- Hyper-V
  - Type 1 hypervisor from Microsoft
  - Architecture
    - Implements isolation of virtual machines in terms of a partition
      - Partition is logical unit of isolation in which each guest OS executes
    - Parent partition
      - Virtualization software runs in parent partition and has direct access to hardware
        Requires supported version of Windows Server
      - There must be at least one parent partition
      - Parent partition creates child partitions which host the guest OSes
        Done via Hyper-V "hypercall" API
      - Parent partitions run a Virtualization Service Provider (VSP) which connects to the VMBus
        Handles device access requests from child partition
    - Child partition
      - Does not have direct access to hardware
        Has virtual view of processor and runs in Guest Virtual Address (not necessarily the entire virtual address space)
      - Hypervisor handles interrupts to processor, and redirects to respective partition
      - Any request to the virtual devices is redirected via the VMBus to the devices in the parent partition
    - VMBus
      - Logical channel which enables inter-partition communication
- KVM (Kernel-based Virtual Machine)
  - Virtualization module in Linux kernel
    - Turns Linux kernel into hypervisor
    - Available in mainline Linux since 2007
  - Can run multiple VMs running unmodified Linux or Windows images
  - Leverages hardware virtualization
    - Via CPU virtualization extensions (Intel VT or AMD-V)
  - But also provides paravirtualization support for Linux/FreeBSD/NetBSD/Windows using VirtIO API
  - Architecture
    - Kernel component
      - Consists of:
        Loadable kernel module, kvm.ko, that provides the core virtualization infrastructure
        Processor specific module, kvm-intel.ko or kvm-amd.ko
    - Userspace component
      - QEMU (Quick Emulator)
        Userland program that does hardware emulation
        Used by KVM for I/O emulations
AWS hypervisor choices & history
- AWS uses custom hardware for faster EC2 VM performance
- Original EC2 technology ran highly customized version of Xen hypervisor
  - VMs can run using either paravirtualization (PV) or hardware virtual machine (HVM)
  - HVM guests are fully virtualized
    - VMs on top of hypervisor are not aware they are sharing with other VMs
  - Memory allocated to guest OSes is scrubbed by hypervisor when it is de-allocated
  - Only AWS admins have access to hypervisors
- AWS found that Xen has many limitations that impede their growth
  - Engineers improved performance by moving parts of software stack to purpose-built hardware components
- C3 instance family (2013)
  - Debut of custom chips in Amazon EC2
    - Custom network interface for faster bandwidth and throughput
- C4 instance family (2015)
  - Offload network virtualization to custom hardware with ASIC optimized for storage services
- C5 instance family (2017)
  - Project Nitro
    - Traditional hypervisors do everything
      - Protect the physical hardware and bios, virtualize the CPU, storage, networking, management tasks
    - Nitro breaks apart those functions, offloading to dedicated hardware and software
    - Replace Xen with a highly optimized KVM hypervisor tightly coupled with an ASIC
    - Very fast VMs approaching performance of bare metal server
- Amazon EC2 – Bare metal instances (2017)
  - Use Project Nitro

Links

End Song
Fax - Stages

For a full transcription of this episode, please visit the episode webpage.

We'd love to hear from you! You can reach us at:

Web: https://mobycast.fm
Voicemail: 844-818-0993
Email: ask@mobycast.fm
Twitter: https://twitter.com/hashtag/mobycast

Activity