2.1 Virtualization — the thing the cloud is built on

The same thing, three names

One idea, sold under a different brand by each provider. Learn it once and you can read any job advertisement.

On AWS

Nitro hypervisor (transparent to you)

On Azure

Hyper-V based

On Google Cloud

KVM based

Work through these

• Hypervisors: type 1 (bare metal) vs type 2 (hosted), and where each is used

  A hypervisor is the layer that lets one physical machine present itself as many, and the two kinds differ in whether they sit directly on the hardware or inside an existing operating system. Cloud providers use the first kind, for reasons this item makes clear.

• Full virtualization, para-virtualization and hardware-assisted virtualization

  Three approaches to the same problem, differing in how much the guest operating system knows it is virtualised and how much help the processor gives. The hardware-assisted approach is why virtualization stopped being slow.

• The VM lifecycle: provision, snapshot, migrate, suspend, destroy; live migration and why it is hard

  A virtual machine can be created, frozen, copied, moved between hosts and destroyed, and moving one while it is running is the hard case. That last capability is what lets a provider service hardware without telling you.

• Containers vs virtual machines — shared kernel, weaker isolation, much faster start

  Containers share the host kernel instead of virtualising a whole machine, which makes them start in moments and isolate less thoroughly. The trade between speed and isolation is the reason both still exist.

• Storage and network virtualization, briefly, and where the noisy-neighbour problem comes from

  Storage and networking are virtualised the same way compute is, and pooling them is where contention appears: another tenant's heavy usage becomes your slow disk. That is what the noisy-neighbour problem means.