Skip to content

Latest commit

 

History

History
93 lines (64 loc) · 3.3 KB

virtual-machines.md

File metadata and controls

93 lines (64 loc) · 3.3 KB
Raw

Overview

Next

Virtual Machines

The focus is on traditional VMs (IaaS) because it provides the utmost flexibility

Language Support : Any

Usage Model : IaaS

Scaling : Server Autoscaling

Primary use case : General workloads - Especially enterprise infrastructure designed to run on servers. This makes those workloads portable and easy to run in the cloud

Compute Engine

Physical servers running in GCP with a variety of configurations.

Important topics covered along with Compute Engine include

  • Machine rightsizing
    • Recommendation engine for optimum machine size
    • Stackdriver statistics
    • New recommendation 24hrs after VM create or resize
  • Global Load balancing
    • Multiple regions for availability
  • Availability policies
    • Live migrate
    • Auto restart
  • Instance metadata and startup scripts
  • Pricing
    • Per second billing
    • Sustained use discounts
    • Committed use discounts
    • Preemptible VMs

Machine Types

Compute Engine provides a variety of machine types; You may also create custom machine types that are tailored to fit you specific use case

Predefined or custom machine types allow you to choose

  • vCPUs (cores) and Memory (RAM)
  • Persistent disks: HDD, SSD, and Local SSD
  • Configure Networking interface
  • Linux or Windows machines

Your choice of CPU will affect your network throughput.

  • Network throughput scales 2Gbps per vCPU core; Except for instances with 2 and 4 vCPUs which receive up to 10Gbps bandwidth.
  • Theoretical max of 32 Gbps with 16vCPUs or 100Gbps with T4 or V100 GPUs
  • 1 vCPU = 1 Hardware hyper-thread

Storage

All Disk types provide the same size options when choosing a persistent disk. So what makes them different?

Performance vs Cost

  • Standard (HDD) persistent disk

                         - Sized up to 64TB/instance.
                     
                     - Performance of disk scales with each GB of space allocated

  • SSD persistent disk

                        - Provide Higher iops/dollar spent

                    - Sized up to 64TB/instance

                    - Performance of disk scales with each GB of space allocated

  • Local SSD (temporary)

                      - Provide even higher throughput and lower latecny because they are attached to the physical hardware.

                  - Persist data until you stop/delete the instance. Typically these disks are used as a swap disk

                  - You can create instances with up to 8 separate 375GB local SSD partitions
                    for a total of 3TB local SSD space for each instance

Networking

  • Default, AM, Custom networks
  • Inbound/Outbound firewall rules - IP based - Instance/group tags
  • Regional HTTPS load balancing
  • Network load balancing - Does not require pre-warming because a load balancer is not a hardware device that needs to analyze your traffic but rather a set of traffic engineering rules that are coming into the Google network and VPC is applying rules destined for your IP address subnet range
  • Global and multi regional subnetworks