Prepare and test your skills

Prepare and test your skills

Scalability and global availability for performance is the ability to keep applications fast during peak demand without wasting resources when demand is low. Google Cloud supports this with elasticity, autoscaling, global region choices, and managed services that reduce operational work.
Elasticity is the ability of cloud resources to grow and shrink as workload demand changes. Autoscaling dynamically adds capacity when demand rises and removes capacity when demand falls, using signals such as CPU utilization, memory usage, and queue depth. Cloud Monitoring (opens in a new tab) and Recommender help find right-sizing opportunities, where right-sizing means changing how much CPU or memory each workload receives to match actual use. Over-provisioning wastes money on idle capacity, while under-provisioning creates bottlenecks and slow performance.
Exam tip: Autoscaling changes how many resources run in response to demand, while right-sizing changes how much resource each workload is allocated to match actual usage.
Google Cloud provides different autoscaling options for virtual machines, containers, and serverless workloads:
| Option | What scales | How it responds to demand |
|---|---|---|
| Compute Engine managed instance groups (MIGs) (opens in a new tab) | A group of VMs managed as one entity | Adds VMs when utilization reaches a high threshold and removes VMs when utilization drops, within a configured minimum and maximum |
| Google Kubernetes Engine (GKE) (opens in a new tab) | Cluster nodes and Pods | Cluster Autoscaler adds or removes nodes, Horizontal Pod Autoscaler changes Pod replicas, Vertical Pod Autoscaler tunes Pod requests and limits, and Node Auto-Provisioning creates optimized node pools |
| Cloud Run (opens in a new tab) | Serverless instances | Scales instances up with incoming traffic, scales down when traffic falls, and scales to zero when there is no traffic or a job completes |
Cloud Run scale-to-zero is especially useful for intermittent or event-driven workloads because instances consume resources only while they actively process requests. Custom metrics can also trigger automated right-sizing actions, and alerts can notify administrators when thresholds are met. Together, these controls help avoid both idle capacity and overloaded systems.
Global infrastructure and availability is the practice of placing workloads and choosing services so users in different locations get responsive access while meeting compliance and cost goals. Google Cloud provides regions around the world, and teams should evaluate available regions for the services they need. Region choice should balance latency for users, compliance requirements for data location, and cost objectives, with help from tools such as the Google Cloud Region Picker. Google-managed services also build on Google's experience running planet-scale, globally distributed systems, which reduces operational burden for customers.
Availability planning also means preparing for disruptions such as natural disasters and cyberattacks. A primary system and its disaster recovery system should both have enough capacity to handle unexpected surges and support graceful failover. This planning connects global placement to capacity decisions, because failover works only when the recovery environment can absorb the redirected load.
Capacity planning, monitoring, and performance testing is the ongoing work of forecasting demand, watching live behavior, and proving the system can handle peaks. Capacity planning analyzes historical usage patterns, growth projections, and business requirements to forecast demand for compute instances, storage, and API requests. Cloud Monitoring and BigQuery collect and analyze usage data such as CPU utilization, memory usage, and network traffic to reveal high-demand periods and potential bottlenecks. Growth projections consider user-base growth, new products and features, marketing campaigns, and service-level agreements and performance targets.
Seasonal variations such as holiday shopping and planned events such as product launches can create temporary spikes that require extra capacity. Continuous monitoring tracks logs that record system events and errors, traces that show how requests flow through an application, and metrics such as round-trip time between services. Common performance-tuning techniques include:
Regular performance testing validates that applications and infrastructure can handle peak loads before users arrive. Load testing simulates realistic traffic patterns, while stress testing pushes the system to its limits to find bottlenecks. Cloud Load Balancing (opens in a new tab) and load testing services help simulate real-world traffic, and teams then adjust virtual machines, containers, or web server and database configurations based on results.
Operational agility and managed service benefits describe how cloud compute lets teams ship applications faster while managing less infrastructure. Virtual machines, containers, and serverless platforms each automate scaling and maintenance in a different way, so teams can choose how much control to keep.
Reduced overhead with managed services means Google Cloud handles routine operational work instead of the customer. Tasks such as patching, scaling, security hardening, and disaster recovery are built into the platform rather than requiring dedicated effort for every workload. This shift lowers long-term operational cost and improves reliability because compliance and reliability features are included by default. Teams then spend more time building applications, which accelerates time-to-market for new features.
Managed virtual machines with MIGs provide VM control with automated group management. Compute Engine managed instance groups (MIGs) (opens in a new tab) are groups of virtual machines managed and scaled as a single entity. An organization defines an autoscaling policy with a minimum and maximum number of VMs and the conditions that trigger scaling. Metrics such as CPU utilization flow into the policy, which adds VMs when utilization reaches a high threshold and removes VMs when utilization drops. This automation provides capacity during peaks without overprovisioning during quiet periods.
Automated container management with GKE removes manual sizing of where containers run. Google Kubernetes Engine (GKE) (opens in a new tab) is a container orchestration platform where the control plane manages placement of containers across cluster nodes. Developers do not manually resize each node for every change in demand because autoscalers adjust resources automatically. GKE uses four complementary tools:
When traffic spikes, Horizontal Pod Autoscaler adds Pods first, then Cluster Autoscaler adds nodes to host those Pods if needed. When demand falls, the process reverses to reduce cost and idle capacity.
Serverless agility with Cloud Run means running code without managing any servers. Cloud Run (opens in a new tab) receives incoming traffic directly and automatically adjusts the number of instances based on that traffic. When traffic increases, Cloud Run scales up instances to handle the load, and when traffic decreases, it scales down instances to reduce costs. Cloud Run can scale to zero when there is no traffic, so resources are powered only while they actively process requests, which is highly effective for intermittent or event-driven workloads.
Exam tip: Cloud Run scales to zero when idle with built-in autoscaling, while MIGs scale between a configured minimum and maximum number of VMs.
Choosing a compute abstraction for agility means matching management level and scaling behavior to the workload. MIGs keep full VM control, GKE optimizes packing and scaling of containers and nodes, and Cloud Run removes infrastructure management entirely:
A decision tree branching on need for VM control and container orchestration to choose between Compute Engine MIGs, GKE, and Cloud Run with their scaling behaviors.
| Option | Management level | Scaling behavior |
|---|---|---|
| Compute Engine MIGs | Customer manages VMs as a group | Adds or removes VMs between minimum and maximum based on CPU or other thresholds |
| GKE | Platform manages nodes and placement, customer manages containers | Adjusts Pods, node size, and node count using four autoscalers |
| Cloud Run | No infrastructure to manage | Automatically scales instances with traffic and scales to zero when idle |
Choose MIGs when the workload needs full VM control with automated group scaling. Choose GKE when containerized applications need efficient packing and flexible scaling of both Pods and nodes. Choose Cloud Run when the goal is fastest time-to-market with pay-only-for-use during requests.
Economic impact of shifting from capital expenditure to operational expenditure is the change from buying hardware upfront to paying only for cloud resources as they are used. With capital expenditure, an organization purchases servers before workloads run, while with operational expenditure the provider owns the hardware and the customer pays under a pay-as-you-go model.
| Model | Upfront investment | How costs occur | Financial risk |
|---|---|---|---|
| CapEx | Large upfront purchase of servers and facilities | Fixed costs for power, cooling, maintenance, and support even when capacity sits idle | High, because money is locked in before demand is known and overprovisioning is needed as a safety net |
| OpEx | No large hardware purchase | Variable costs based on actual use of virtual machines, containers, and serverless instances | Lower, because resources can be added or removed dynamically and spending tracks business activity |
The table shows why cloud compute changes budgeting. CapEx commits funds to peak capacity, while OpEx lets an organization start small and scale resources dynamically to meet organizational goals.
Capital expenditure (CapEx) is upfront spending to own physical IT assets such as servers, storage, networking, and data center space. A total cost of ownership model for CapEx includes hardware and software licenses plus ongoing costs for power, cooling, maintenance, network use, and support services. Because on-premises capacity is rigid and takes time to add, teams often overprovision extra servers to handle seasonal peaks, product launches, and failover. That safety-net capacity sits unused most of the time but still costs money to own and operate.
Operational expenditure (OpEx) is ongoing spending to use cloud resources without owning the underlying hardware. An organization provisions a virtual machine, container cluster, or serverless instance when a workload needs it and releases it when demand drops. Workload metrics such as CPU utilization, memory usage, or incoming traffic flow into autoscaling actions that add capacity during peaks and remove capacity when utilization falls. Compute Engine (opens in a new tab) managed instance groups, Google Kubernetes Engine (GKE) (opens in a new tab), and Cloud Run (opens in a new tab) all provide autoscaling, so applications keep the resources they need without paying for idle safety-net capacity.
Flexible pricing that makes OpEx controllable is a set of billing options that lowers the rate when usage is predictable or flexible. The options let an organization match price to workload pattern instead of paying one fixed on-demand rate for everything. Steady workloads earn discounts while short or interruptible work uses low-cost spare capacity, which reduces financial risk.
OpEx still requires visibility to stay efficient. Tags and labels attach business metadata such as project, team, or department to resources, so spending flows from each resource to its owner in billing reports. Cloud Billing (opens in a new tab) budgets and alerts track actual spending against planned limits and notify stakeholders when budgets are exceeded, which supports accountability and helps prevent uncontrolled growth.
Exam tip: Budgets and alerts notify stakeholders of potential overruns but do not automatically stop resources — teams must still take corrective action.
Total cost of ownership (TCO) and business value compare the full cost of on-premises CapEx plus operations against cloud OpEx over time. A TCO model makes hidden on-premises costs visible, including power, cooling, maintenance, and idle overprovisioned capacity. Elastic scalability makes it easier to reduce costs in the cloud than in a rigid data center because resources can be right-sized and automated. The business value is lower upfront investment, improved operational efficiency from automation, and the ability to direct savings toward innovation rather than maintaining idle hardware.
Autoscaling changes how many resources run in response to demand, while right-sizing changes how much CPU or memory each workload is allocated to match actual use. Autoscaling adds capacity when demand rises and removes capacity when demand falls, which helps avoid both idle capacity and overloaded systems.
Choose Cloud Run when the goal is fastest time-to-market with pay-only-for-use during requests because it automatically scales instances with traffic and scales to zero when idle. Compute Engine managed instance groups (MIGs) keep full virtual machine control with scaling between a configured minimum and maximum, while Google Kubernetes Engine (GKE) optimizes packing and scaling of containers and nodes. Cloud Run is highly effective for intermittent or event-driven workloads because instances consume resources only while they actively process requests.
Capital expenditure (CapEx) is upfront spending to own physical IT assets such as servers and data center space with fixed costs for power, cooling, maintenance, and support even when capacity sits idle. Operational expenditure (OpEx) is ongoing spending to use cloud resources without owning the underlying hardware, with variable pay-as-you-go costs that rise and fall with actual use of virtual machines, containers, and serverless instances.
No, Cloud Billing budgets and alerts notify stakeholders when spending exceeds planned limits but do not automatically stop resources. Teams must still take corrective action to address potential overruns and prevent uncontrolled growth.
A global e-commerce retailer is migrating its customer-facing web application to Google Cloud. During seasonal sales and promotional events, the application experiences massive spikes in traffic from shoppers worldwide, followed by extended periods of low activity. The company wants to achieve the following business and technical goals:
How do Google Cloud's global infrastructure and elasticity capabilities work together to meet these requirements?
Resource quotas cap the maximum number of compute cores provisioned across regions, while production code freezes prevent system changes during seasonal peak periods.
Infrastructure as Code (IaC) deploys immutable virtual machines in a single central region, while administrators manually redeploy the environment whenever traffic volume changes.
Code profiling and database indexing optimize software execution speed, while a fixed number of high-capacity virtual machines run continuously to absorb peak loads.
Multi-region deployments with global load balancing route requests to the closest healthy region, while autoscaling dynamically adds compute resources during traffic spikes and removes them when demand drops.
Last updated