Google Compute Engine

Google Compute Engine. It is an Infrastructure as a service (IaaS) component of the Google Cloud Platform, allowing users to launch and manage virtual machines on Google's infrastructure. The service was first introduced in limited preview at Google I/O in 2012 before becoming generally available in 2013. It provides scalable, high-performance VMs that run on the same infrastructure used for services like Google Search and YouTube.

Overview

As a core offering of the Google Cloud Platform, it provides raw compute resources in a secure, global environment. The service operates across Google's extensive network of data centers and regions, such as those in Iowa and Belgium, ensuring low-latency access worldwide. It integrates deeply with other Google Cloud Platform services like Google Cloud Storage and Google Kubernetes Engine for comprehensive cloud solutions. Customers range from startups to large enterprises like Spotify and Snap Inc., leveraging its infrastructure for diverse workloads.

Features

Key capabilities include custom machine types, allowing users to tailor vCPU and RAM configurations beyond predefined types. It offers both Standard Persistent Disk and high-performance SSD Persistent Disk for block storage, with automatic encryption. Features like Live Migration ensure VMs are maintained during host events without downtime. The service supports a wide array of operating systems, including public images for Red Hat Enterprise Linux, SUSE Linux Enterprise Server, and Windows Server. Advanced networking features encompass Virtual Private Cloud, Cloud Load Balancing, and Cloud CDN for optimal performance.

Architecture

The underlying infrastructure is built on Andromeda, Google's software-defined networking stack that enables high-performance network virtualization. VMs run on custom hardware, including TPUs for machine learning and the Intel Skylake platform for general compute. Global resources are organized hierarchically into regions and zones, facilitating high availability and disaster recovery designs. The control plane is managed by Borg, Google's internal cluster management system, which also orchestrates services like Gmail. Storage is decoupled from compute, with Persistent Disk offering durable block storage.

Pricing

It employs a sustained-use discount model, automatically applying lower rates to VM instances that run for a significant portion of a billing month. Customers can also purchase Committed Use Discounts for predictable workloads or use Preemptible VMs for fault-tolerant jobs at a substantially reduced cost. Pricing varies by factors like machine type, selected region such as Tokyo or South Carolina, and attached Persistent Disk storage. Network egress pricing follows a tiered structure, with charges decreasing as data volume increases, and ingress is typically free.

Use cases

Common applications include running large-scale Apache Hadoop or Apache Spark clusters for data processing and analytics. It is extensively used for hosting complex, multi-tier web applications, often integrated with managed databases like Cloud SQL. The platform supports high-performance computing workloads, such as financial modeling and genomic sequencing, leveraging its powerful VMs. Media companies like The New York Times have used it for rendering and batch processing, while research institutions utilize it for scientific simulations.

Comparison with other services

Compared to Amazon EC2 from Amazon Web Services, it is often noted for its per-second billing, sustained-use discounts, and deep integration with Google Kubernetes Engine. Versus Microsoft Azure Virtual Machines, it differentiates with its global Andromeda network backbone and custom machine types. It generally offers fewer instance family variations than EC2 but provides strong performance consistency. The service's approach to Preemptible VMs is analogous to AWS Spot Instances but with a fixed maximum runtime. Its container ecosystem, centered on Kubernetes, contrasts with AWS Fargate's serverless container model.