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)
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:
