Day 2 - Virtualization

📌 Key Takeaways

• Virtualization is a technology for running multiple operating systems at the same time on one physical computer.
• Full virtualization does not require modifying the guest OS, while paravirtualization requires modifying the guest OS.
• Hypervisors come in two types — Type 1 (directly on the hardware) and Type 2 (on top of an operating system).
• VirtualBox is Oracle's Type 2 product, released under the GPL, and uses full virtualization.
• VirtualBox supports VMware's VMDK and Microsoft's VHD images.

What Is Virtualization

• The hypervisor divides up the hardware and allocates it to each virtual machine.
• The physical computer is called the host, and the virtual computers on it are called guests.
• A hypervisor is also called a VMM (Virtual Machine Monitor). It is another name for the same thing.

• A guest's disk is ultimately a file on the host. That is why virtual machines are easy to copy, back up, and move, and the format of this file (VDI, VMDK, VHD) differs by product, which is why it appears on the exam.

• Snapshots are the reason we recommended "install it in a virtual machine if it's for study" in week 2. Even if you partition wrongly, you can just roll back.

The Two Types of Hypervisors

• Type 1 acts as an operating system itself. That is why it is fast.
• Type 2 is installed like a program on top of an already installed operating system. It is easy to install, so it is used for personal use.

      Type 1 (native)                   Type 2 (hosted)
 
   Guest OS    Guest OS             Guest OS    Guest OS
   ─────────────────────            ─────────────────────
       Hypervisor                       Hypervisor      ← just one more program
   ─────────────────────            ─────────────────────
        Hardware                        Host OS         ← one extra layer
                                    ─────────────────────
                                         Hardware

• The deciding question is "what is underneath the hypervisor?" Hardware means Type 1; an operating system means Type 2.
• The reason Type 2 is slower is the one extra layer in the diagram. A guest's request goes through the hypervisor and then through the host OS once more before reaching the hardware.
• KVM is where people get confused. You install Linux first and then use it, so it looks like Type 2, but once the module is loaded, the Linux kernel itself acts as the hypervisor, so it is classified as Type 1. The exam follows this classification.
• Products from the same company exist on both sides — VMware ESXi is Type 1, VMware Workstation is Type 2. Choosing by the name "VMware" alone will get you the wrong answer.

Full Virtualization and Paravirtualization

The deciding criterion is "must the guest OS be modified?" Full virtualization works as is, without modification, while paravirtualization requires modification. In exchange, paravirtualization is faster.

• Intel VT-x and AMD-V are CPU features that assist virtualization. Thanks to them, full virtualization's performance improved greatly, and full virtualization is now the mainstream.
• Xen started out with paravirtualization but now supports full virtualization as well.

Why Must the Guest Be Modified

Operating systems are built on the belief that they have the hardware all to themselves. When several such operating systems run on one computer, someone has to direct traffic.

   Full virtualization                   Paravirtualization
 
   Guest OS: "I write to the disk        Guest OS: "Hypervisor, please write this"
        directly"                             │  (knows it is a virtual machine)
        │  (doesn't know it is a VM)          ▼
        ▼                                Requests directly via a hypercall
   The hypervisor intercepts the command  → no intercepting or imitating step
   and imitates the real hardware to
   handle it

• The third row is why paravirtualization, despite being fast, did not become the mainstream. An operating system whose source cannot be modified cannot be run.
• As the CPU began to help directly with the interception (Intel VT-x, AMD-V), full virtualization's weakness shrank. This approach of getting help from the CPU is called hardware-assisted virtualization.

On Linux, you check whether the CPU has this feature in /proc/cpuinfo.

grep -E -c 'vmx|svm' /proc/cpuinfo
# 4

• The printed number is the count of logical CPUs where the flag appears. The example above means all 4 cores support it, and 0 means hardware-assisted virtualization cannot be used in that environment.
• Questions swap the flag name and the company. vmx is Intel, svm is AMD.

Features by Product

VirtualBox

This is an item that appears on the exam exactly as the product spec.

⚠️ The statement "VirtualBox supports paravirtualization" is presented as the wrong one. VirtualBox is a full virtualization product that uses Intel VT-x and AMD-V. On the other hand, the statement that it supports VMDK and VHD images is correct.

• The reason choices say "the base package is GPL" rather than "all of VirtualBox is GPL" is the Extension Pack.
• The actual VirtualBox settings window has an item called "paravirtualization interface", which can be confusing. This is a helper feature that tells the guest things like timing information, and the product's virtualization classification is full virtualization. On the exam, answer according to the classification.

Virtual Disk Image Formats

This is the material for questions that pair products with formats.

• VirtualBox doesn't only use its own format (VDI); it also reads other vendors' formats (VMDK, VHD). A choice saying "it does not support images from other companies" is a wrong statement.
• VDI has two meanings. In the VirtualBox context it is a disk image format; in the technology trends context it is desktop virtualization (Virtual Desktop Infrastructure — a way of receiving a desktop screen kept on a server remotely). You tell them apart from the context of the passage.

VMware

• VMware being a commercial product is what sets it apart from the open-source VirtualBox (base package), KVM, and Xen.

Xen

• Xen calls virtual machines domains. Dom0 is for management, and DomU is for regular guests.
• It is asked as "the representative of paravirtualization", but the statement "it cannot do full virtualization at all" is wrong.

KVM

• You don't install a separate hypervisor; once you load the kernel module, Linux itself becomes the hypervisor. That is the definition of KVM.

lsmod | grep kvm
# kvm_intel             368640  0
# kvm                  1028096  1 kvm_intel
virsh list --all
#  Id   Name    State
# ------------------------
#  1    web01   running
#  -    db01    shut off

• kvm is the common module, and kvm_intel (Intel) or kvm_amd (AMD) is loaded alongside it to match the CPU. The 1 kvm_intel at the end of the kvm line means "this module is being used by one module, kvm_intel".
• virsh list --all shows even virtual machines that are powered off. Only running ones get an Id number; stopped ones show -.
• KVM is a full virtualization method that only works when the CPU has VT-x / AMD-V. It has been included in the kernel since Linux kernel 2.6.20.

• QEMU on its own is an emulator that even imitates other CPU architectures, so it is slow. Used together with KVM, the real CPU executes the CPU instructions directly, so it gets fast. That is why the two always go together.

Containers — Different from Virtualization

• A container does not run a separate operating system. It shares the kernel and only isolates processes. That is why it is light and fast.
• Docker is not a virtual machine. Questions asking for this distinction appear.

      Virtual machine                    Container
 
   App        App                      App        App
   Guest OS   Guest OS                 ─────────────────
   ─────────────────                   Container engine (Docker)
     Hypervisor                        ─────────────────
   ─────────────────                    Host OS (one kernel shared by all)
       Hardware                        ─────────────────
                                           Hardware

• In the diagram, the whole guest OS layer is missing. That is the entire reason containers are light and start quickly.
• Containers are built from two Linux kernel features — namespaces, which divide what a process can see, and cgroups, which limit the resources it can use.
• Because the kernel is shared, you cannot run an operating system with a kernel different from the host's. Inside a container on a Linux host it is always Linux. To run Windows, you need a virtual machine.

Telling Them Apart from Passages

Commonly Missed Points

1. Paravirtualization is the one that doesn't modify the guest → The opposite. The one that doesn't modify it is full virtualization. Remember "full" as imitating the hardware completely.
2. Full virtualization is always faster → In principle, paravirtualization is faster. Full virtualization only narrowed the gap with CPU support.
3. VirtualBox is Type 1 → It is Type 2, installed on top of an operating system.
4. KVM is Type 2 → The kernel itself becomes the hypervisor, so it is Type 1.
5. All VMware products are Type 2 → ESXi is Type 1.
6. VirtualBox only uses the VDI format → It also supports VMDK and VHD.
7. VHD is VMware's format → VHD is Microsoft's; VMware's is VMDK.
8. vmx is AMD's flag → vmx is Intel, svm is AMD.
9. Docker is a Type 2 hypervisor → It is not a hypervisor but container technology.
10. Each container has its own kernel → They share one host kernel. The ones with separate kernels are virtual machines.

Tomorrow we cover embedded systems and technology trends.

📖 Glossary

• Virtualization : A technology for running several operating systems at the same time on one physical computer.
• Hypervisor : Software that divides hardware resources among virtual machines.
• Type 1 / Type 2 : Running directly on the hardware / being installed on top of an operating system.
• Full virtualization : A method that imitates the hardware without modifying the guest OS.
• Paravirtualization : A method that modifies the guest OS so it requests directly from the hypervisor.
• VirtualBox : Oracle's Type 2 virtualization product. It is released under the GPL and uses full virtualization.
• KVM : A virtualization technology that turns the Linux kernel itself into a hypervisor.
• Container : A lightweight method that shares the host kernel and only isolates processes. Docker is the representative example.