Virtualization and CPU Organization Basics

### Guard Application

• A new instance of Windows runs at the hardware layer, with an independent copy of the kernel and Windows Platform Services
• Microsoft Edge is run by the instance
• The hardware restricts access to the user's normal operating environment

Type 2 Hypervisor

• Virtual Machine Monitors (VMMs) are built on top of a host operating system
• The host operating system handles resource allocation and provides execution environments for guest operating systems
• User-mode Linux (UML) and QEMU are examples of Type 2 Hypervisors

Virtualization Methods

• Popek-Goldberg VMMs
• Dynamic Binary Translation
• Hardware-assisted Virtualization

CPU Organization Basics

• Instruction Set Architecture (ISA) describes the state visible to the programmer: registers and memory, and instructions that operate on the state
• ISA is typically divided into two parts: User ISA and System ISA

### User ISA

• The part of ISA that is mostly used for computation
• The State includes registers and memory

### System ISA

• The part of ISA that is mostly used for system resource management
• Privilege levels, control registers, traps and interrupts, MMU (including page tables and Translation Lookaside Buffer (TLB), and I/O device access are part of the System ISA

### Virtualization with Interpretation

• Fetch/Decode/Execute pipeline is executed in software
• Instructions are converted to the minimal set of host instructions
• Designed for use by debuggers to set breakpoints

Disco

• Disco operating system allows multiple copies of the Silicon Graphics IRIX operating system to run on Stanford Flash Multiprocessors
• Disco was built to run on shared memory multiprocessors
• The Stanford Flash is a cache coherent NUMA (Non-Uniform Memory Access) system
• Disco evolved into VMWare, built by the same creators

VMWare

• The initial product was to help developers with testing
• The "killer" product was server consolidation in enterprise environments
• Server consolidation resulted in cost reduction and better management
• VMWare now generates over 90% of its revenue from server consolidation

Virtualization with Dynamic Binary Translation

• Guest instructions are translated into host instructions
• Translation caching reduces overhead
• Still slower than direct execution
• QEMU performs this type of translation

### Namespaces

• Namespaces allow for process isolation
• Examples include network isolation, process IDs, mounting, and inter-process communication

From Virtual Machines to Linux Containers

• Cgroups and namespaces can be used to create containerized processes
• LXC (Linux Containers) is user-land software providing isolated Linux environments on a single host
• LXC can be used to run multiple Linux distributions as containers

From LXC to Docker

• Docker was built to address developer needs
• Docker supports container image formats and building, managing, sharing, and running container images using Dockerfile, docker build, docker images, docker rm, docker ps, docker push/pull, and docker run
• Docker provided solutions for moving images, copy-on-write, sharing previously created images
• Docker addresses dependencies and configurations when building, testing, and deploying applications across servers by packaging application and dependencies as container images

### Container Ecosystem Today

• Low-level container runtimes manage namespaces, control groups, and container execution (e.g., runC, lxc, rkt)
• High-level container runtimes handle image formats, image management, and image sharing (e.g., cri-o, containerd)
• Docker is a collection of components, including the Docker engine (a user-facing daemon, REST API, and CLI) and a runtime (containerd, container-shim, and runC)

### Advantages of containers

• Abstraction levels: Hypervisors work with hardware abstraction while containers work at the OS abstraction level
• Density: Containers require less space than virtual machines
• Elasticity: Easier to scale containers than virtual machines
• Improved software development lifecycle: Building, testing, and deploying are made easier with containers

### Serverless Functions

• Serverless function handlers run in response to events including web requests, database updates, and scheduled events.
• Function invocations are based on a pay-per-use model where billing depends on the time and memory used during each function invocation.
• Server pool sharing with customers allows for efficient utilization of resources.
• Serverless functions encourage specific runtimes for minimal network copying and code residency in memory.

Architecture

• Architecture is characterized by a function request, function registry, and scaling engine. The function registry contains the function code, while the scaling engine controls the deployment and scaling of functions.

Functions vs Containers

• Experimental setup used Elastic Beanstalk (autoscaling cloud service) and AWS Lambda (serverless functions for AWS)
• Workload simulation uses 100 concurrent HTTP requests and 200 ms compute time per request for one minute
• AWS Lambda RPC has a median response time of 1.6 seconds due to efficient startup of 100 worker instances, compared to 20 seconds for Elastic Beanstalk.

### Scalability Result

• AWS Lambda RPC has a median response time of only 1.6s.
• Lambda was able to start 100 unique worker instances within 1.6s.
• An RPC in Elastic Beanstalk often takes 20s.

Three Generations of Virtualization

• There are three generations of virtual machines, the latest includes serverless computing.

This quiz covers key concepts in virtualization, including Guard Applications, Type 2 Hypervisors, and various virtualization methods. Additionally, it explores the fundamentals of CPU organization and the Instruction Set Architecture (ISA), distinguishing between User ISA and System ISA. Test your understanding of these critical components in computer systems.