ITSS 4370 - Module 4 - Virtualization

Server Virtualization

The hypervisor, is a low-level program that allocates hardware resources such as CPU, memory, hard-disk, etc. to the VMs (guests) in real time.

Type 1 Hypervisor

Referred to as “bare metal”

• Efficient and scalable

• Secure

• Used in production environments

• Examples: VMware, ESX/ESXi, Microsoft Hyper-V, Xen releases

• Does not include operating system

Type 2 Hypervisor

• Latency issues

• Great for personal or development purposes

• Examples: VMware, Workstation/Fusion/Player, Oracle Virtualbox

• Does include an operating system

Why use Server Virtualization?

• Underutilized Hardware

  • Traditional data centers plan and provision servers to handle peak loads. But, workloads change throughout the day, week, month, and year.

  • Server utilizalization is around 5-10%

• High Data center Space Costs

  • Physical servers housed in the data centers consume space (even when idle)

• High Energy Costs

  • Power and cooling costs money

• High Costs of IT operations

  • Installation, patching, upgrades

Demand fluctuates over time

Benefits of Virtualization

• Server consolidation reduces the data center footprint and lowers energy costs, cost of IT operations, etc, i.e., reduces TCO

• Virtualization improves utilization rates to 70-80%

• Rapid elastic compute capacity - VMs can be launched quickly

• High availability

• Minimized downtime

Example: Instead of running 150 physical servers, you can run 150 VMs on 15 physical servers

Planning for Virtual Machines

What is overcommitment? - Assigning more resources to VMs than available physical resources on the host

• A physical server may run diverse applications with different requirements

• Critical Components - Compute (CPU), memroy (RAM), and disk space

• It is a common practice to overprovision these resources to maximize the workloads (applications) on a host

  • Choosing not to overprovision is safe but wasteful. Too much overcommitment

CPU

Processing power is assigned to virtual machine as virtual CPUS (vCPUs). The vCPUs are allocated from the host’s available physical CPUs (pCPUs).

Guidelines for CPU overcommitment:

• Overcommitment ratio (vCPU to pCPU) should be determined based on the type of workloads you are running. A higher ratio is not recommended if the host is running all workloads that are process-intensive.

• It is safe to maintain an overcommitment ratio of 3:1 i.e., each pCPU supports three vCPUs.

• Close monitoring of metrics becomes important for higher ratios such as 4:1 or 6:1

Memory

• Over commitment ratio is 1.25:1 is safe

• Following measures are useful to maximize the use of RAm

Memory Ballooning

• The hypervisor reclaims memory from other VMs

• Places all the loaned memory pages into a balloon that is made available to the VM that urgently needs it

Transparent Page Sharing

• Deduplication technique, minimal performance impact

Memory Compression

• Memory pages are compressed and placed into a compression cache on disk. When needed, a page is retrieved from the cache and uncompressed. Some performance impact.

Swapping or Paging to Disk

• When insufficient memory is available, the hypervisor will dump data from memory to a page file on the hard disk. Retrieval from disk is slow regarding the overall performance.

• Used as a last resort.

Storage

• A virtual disk is utilized by a virtual machine to store the operating system, application software, and other files that it needs

• There are two ways virtual disks can be provisioned to virtual machines

  • Thick (Fixed) Provisioning - The size is specified and pre-allocated during creation

  • Thin (Dynamic) Provisioning - Starts small, additional chunks are allocated as needed

• Thin provisioning has shown to have negligible impact on performance and is the widely accepted method for overcommitting storage

Virtual Machine Cloning

• “Copy” or “replica” of an existing VM

• The existing VM is called the parent and the new VM is the clone

• There are two types of clones: full and linked. A linked clone shares virtual disks with the parent, and a full clone creates its own. A full clone has faster performance but takes longer to create.

• Cloning allows faster deployment of identical VMs to a group.

• Use Case: Setting up a development or a testing environment

Virtual Machine Snapshots

• Captures the state of a VM at the specific time when the snapshot is taken

• What is contained in a snapshot?

  • State of the VM’s disks

  • Contents of the VM’s memory

  • VM settings

• Can be taken when a machine is running, powered off, or suspended

  • Best practice: Take a snapshot before a major software installation or maintenance task. In case of failure or issues, the machine can be restored back to its 'previous’ state in a short time

What is a container?

• A __________ is a unit of software that packages all its code and dependencies so that the application runs consistently in any environment.

• A runtime engine such as docker enables all containers on a host to share the host OS kernel.

• A container has a docker engine that allows for containers on top of the host

  • No operating systems in their containers, they share it with the kernel

Virtual Machines vs. Containers

VMs

• Each VM runs in its own OS

• Heavyweight (several GBs in size)

• Hardware-level virtualization

• Startup time in minutes

• Fully isolated and hence more secure

Containers

• All containers share the host OS

• Lightweight (tens of MBs in size)

• OS virtualization

• Startup time in milliseconds

• Possibly less secure

Monolith vs. Microservices-based Application

Monolith:

• Has to go sequentially

• Coupled together

Micro-serviced:

• Removes any dependencies from each other and each business function now runs as a separate service

• Run through API calls

• Benefits: Can assign multiple teams and work on one service, can be done in different languages,

What are Microservices?

• Common to cloud-native applications

• A microservice-based application is composed of multiple loosely coupled and independently deployable components called as “Services”.

• Services are deployed over containers

• Challenges:

Intro to Serverless Computing

• Serverless means that cloud consumers are free from the administrative burden of provisioning and managing servers as that responsibility is handled by the CSPs.

• AWS Example: AWS Lambda

• Application developers write code to AWS lambda and do not have to worry about the infrastructure or scale required for the application. A multitude of programming languages such as Java, Python, Ruby, etc, are supported.

• Serverless applications are triggered by events.

• You only pay for compute when your code runs

• Benefits: Lower costs, increase in productivity, better operational efficiency and simplified management