GCP - Google Cloud Computing - Compute Engine

Posted Jan 1, 2021 Updated Aug 7, 2024

By Grace L

5 min read

Google Cloud Computing - Compute Engine

Compute Engine

Infrastructure as a Service
A managed environment
for deploying virtual machines
- run thousands of virtual CPUs on a system that is designed to be fast and to offer consistent performance.
Fully customizable VMs
- Compute Engine offers virtual machines that run on GCP
- create and run virtual machines on Google infrastructure.
- run virtual machines on demand in the Cloud.
  - select predefined VM configurations
  - create customized configurations
no upfront investments
Lift and shift
choice:
- have complete control over your infrastructure
  - maximum flexibility
  - for people who prefer to manage those server instances themselves.
  - customize operating systems and even run applications that rely on a mix of operating systems.
- best option if other computing options don't support your applications or requirements
  - easily lift and shift your on-premises workloads into GCP without rewriting the applications or making any changes.

SSD

standard, SSD, or local SSD.
- the standard spinning hard disk drives or HDDs, or flash memory solid state drives SSDs.
- Both of these options provide the same amount of capacity in terms of disk size when choosing a persistent disk.
Therefore, the question really is about performance versus cost
standard disks
- higher amount of capacity for your dollar.
SSDs
- higher number of IOPS per dollar
local SSDs
- even higher throughput and lower latency than SSD persistent disks because they’re attached to the physical hardware.
- but data on local SSD persists only until stop or delete the instance.
- Typically is used as a swap disk just like a RAM disc.
- But if you need more capacity, you can store those on a local SSD.
one instances with up to 8 separate 375 GB local SSD partitions for total of 3 TB of local SSD space for each instance.
Standard and non-local SSD disks can be sized up to 64 TB for each instance. The performance of these disks scales with each GB of space allocated.

Linux and Windows Server images provided by Google or customized versions of these images
import images for many of the physical servers.

Machine type OS

Linux and Windows Server images provided by Google or customized versions of these images
import images for many of the physical servers.

Machine type

Processing power

machine learning and data processing that can take advantage of GPUs, many GCP zones have GPU’s available for you.
Just like physical computers need disks, so do VM.

Virtual machines need block storage

(2 main choices)
1. 2 persistent storage
- standard or SSD.
- offer network stores that can scale up to 64 terabytes
- can easily take snapshots of the disks for backup and mobility
  1. attach a local SSD
- If application needs high-performance scratch space
- enable very high input/output operations per second
to store data of permanent value somewhere else
- local SSDs content doesn’t last past when the VM terminates.
- persistent disks can.

boot image.

choose a boot image.
- GCP offers lots of versions of Linux and Windows
- can import the own images too.

VM startup scripts

let the VMs come up with certain configurations
- like installing software packages on first boot.
pass GCP VM startup scripts to do so.
- can also pass in other kinds of metadata too.

snapshot

grained control of costs

per second billing
- GCP enables fine grained control of costs of Compute Engine resources by providing per second billing.
- helps reduce the costs when deploying compute resources for short periods of time
  - such as batch processing jobs

preemptible VMs instances 抢先的

provide significantly cheaper pricing for the workloads that can be interrupted safely
- have a workload that no human being is sitting around waiting to finish
- such as batch job analyzing large dataset
benefit:
- save money
- cost less per hour but can be terminated by Google Cloud at any time.
different from an ordinary Compute Engine VM in only one respect.
- given compute engine permission to terminate it if it’s resources are needed elsewhere.
- make sure the job able to be stopped and restarted.
can’t convert a non-preemptible instance into a preemptible one.
- must be made at VM creation.
choose the machine properties of the instances
- such as the number of virtual CPUs and the amount of memory
- by using a set of predefined machine types
- or by creating the own custom machine types.
- the maximum number of virtual CPUs and the VM was 96 and the maximum memory size was in beta at 624 gigabytes.
- huge VMs are great for workloads like in-memory databases and CPU intensive analytics
- but most GCP customers start off with scaling out not scaling up.

Auto scaling

add and take away VMs from the application based on load metrics.
place the Compute Engine workloads behind global load balancers that support autoscaling
balancing the incoming traffic across the VMs
- Google VPC supports several different kinds of load balancing

managed instance groups

Availability policies

If a VM is stopped (outage or a hardware failure), the automatic restart feature starts it back up. Is this the behavior you want? Are the applications idempotent (written to handle a second startup properly)?
During host maintenance, the VM is set for live migration. However, you can have the VM terminated instead of migrated.

bills by the second
- with a one-minute minimum for virtual machine usage.
Sustained-use discounts
- apply automatically to virtual machines that run for more than 25% of a month.
Committed-use discounts
- for stable and predictable workloads.
Preemptible and Spot VMs
- can be used to save money for certain workloads
- Preemptible and spot VMs have the same performance as ordinary VMs of the same machine type.
- Preemptible 可優先的 VMs can run for up to 24 hours, while Spot VMs have no maximum runtime.