Prepare and test your skills

Prepare and test your skills

Distinguishing between zonal and regional Persistent Disks means choosing between one copy of data in one zone and two synchronously replicated copies in two zones. Both options use Persistent Disk (opens in a new tab) as durable network block storage that attaches to a Compute Engine virtual machine (VM) over the network. The choice trades lower cost and simplicity against higher availability and faster recovery from a zonal outage.
Zonal Persistent Disk is block storage that exists in only one zone, such as us-west1-a. The VM that uses the disk must run in that same zone, so compute, data, and requests stay together for low latency. Compute Engine still protects the disk inside that zone by spreading data across several physical disks, calculating checksums on every operation to verify reads, and maintaining availability through maintenance events.
Zonal disks fit general boot and data use when an application can be rebuilt from backup if its zone fails. Protection against user error comes from incremental snapshots, which can be created in minutes even while the disk is attached to a running VM. Performance scales predictably with provisioned capacity and with the number of VM vCPUs until the VM limit is reached, and new volumes reach full performance immediately without pre-warming or striping.
Regional Persistent Disk is block storage with the same basic qualities as zonal disk, plus synchronous replication between two zones in the same region. When an application writes data, the write flows from the VM to the replica in the local zone where the VM runs, such as us-west1-a, and then is synchronously replicated to the second zone, such as us-west1-b. One of the two replication zones must be the same zone where the VM runs.
This two-zone layout protects mission-critical data if one zone fails, because the surviving replica in the other zone remains available. If a zonal outage occurs, the workload can usually fail over to a VM in the second zone and continue using the same regional disk. If both replication zones fail, the event is treated as a region-wide failure.
Exam tip: Regional replication keeps the disk available in the surviving zone, but a zonal outage can still stop the original VM, so failover still requires a working VM in the surviving zone.
Choosing between zonal and regional disks depends on how much availability, durability, and recovery speed the workload needs versus write performance and flexibility. Regional disks provide about twice as many disk replicas as zonal disks and lower Recovery Point Objective (RPO) and Recovery Time Objective (RTO) than snapshot restore. Regional disks are designed to work with regional managed instance groups, which distribute VMs across zones in a region.
| Option | Storage scope | Availability and recovery |
|---|---|---|
| Zonal Persistent Disk | One copy in one zone; VM and disk must share that zone | Protects against equipment failure inside the zone; recovery from zone loss depends on snapshots, with higher RPO and RTO |
| Regional Persistent Disk | Two synchronously replicated copies in two zones of one region; one replica must share the VM zone | Survives loss of one of the two zones with immediate access to the second replica; provides lower RPO and RTO than snapshots |
Choose Regional Persistent Disk when redundancy across zones matters more than maximum write performance, such as for databases and high-availability services. Choose zonal disk when cost, broader machine-type support, and simpler operation matter more, and when snapshot-based backup meets recovery needs.
Durability by design also differs by disk type and scope:
| Disk type | Zonal durability | Regional durability |
|---|---|---|
| Standard | Better than 99.99% | Better than 99.999% |
| Balanced | Better than 99.999% | Better than 99.9999% |
| SSD Performance | Better than 99.999% | Better than 99.9999% |
Regional Persistent Disk performance follows the same scaling model as zonal disk, where greater provisioned capacity and more VM vCPUs allow greater input/output operations per second (IOPS) and throughput. New regional volumes reach maximum performance immediately without pre-warming, and an administrator can resize a regional disk larger to add space, throughput, and IOPS. Write operations still travel over the network and count toward the VM network egress cap, and regional replication adds cross-zone write work, so regional disks are preferred when redundancy matters more than peak write speed.
Reliability for Regional Persistent Disk comes from two layers of redundancy. Compute Engine replicates disk data to the two selected zones, and within each zone the data of each replica is spread across multiple physical machines. For robust systems, combine Regional Persistent Disk with incremental snapshots for protection against accidental deletion and with regional managed instance groups so replacement VMs can start in another zone.
Regional Persistent Disk limits affect VM selection, disk creation, and resizing, so they must be planned before choosing regional storage. These constraints control which machine types can attach the disk, how the disk can be created, and how it can grow.
Aligning storage characteristics with workload requirements means matching persistence, availability, and performance controls to what the workload actually needs. For Compute Engine (opens in a new tab) VMs, the choice centers on Persistent Disk and Google Cloud Hyperdisk (Hyperdisk), plus snapshots and backup storage for durability. Zonal options cost less and suit general tiers, while synchronously replicated options cost more but keep data available through a zonal outage.
Persistent Disk is durable network block storage that a Compute Engine instance can access over the network. The disk is independent of the instance, so it can be detached and moved and its data is preserved even if the instance is stopped or deleted. Persistent Disk has built-in redundancy to protect data against equipment failure and to keep data available during data center maintenance events.
Successful completion of a write occurs only after data has been transferred to stable media, because Persistent Disk volumes do not cache writes in a way that risks loss. Data on Hyperdisk volumes is also preserved even if the instance is stopped. Hyperdisk volumes offer configurable performance and can be resized dynamically without recreating the workload, while Persistent Disk remains the durable choice when a machine series does not support Hyperdisk.
Zonal Persistent Disk is a Persistent Disk volume that lives in a single zone and is a practical default for application and web tiers. The reference architecture uses zonal Persistent Disk volumes for all tiers when low cost and single-zone persistence are enough. Performance scales with the size of the volume, so a larger zonal disk provides more throughput and IOPS for the same disk family.
Regional Persistent Disk is a Persistent Disk volume that replicates writes synchronously across two zones in the same region. In the event of a zonal outage, a backup VM in the secondary zone can force-attach the regional disk and continue serving data without a restore. Choose Regional Persistent Disk when the workload must remain available if one zone fails, such as a self-managed database that cannot tolerate rebuilding storage from backup.
Hyperdisk Balanced provides better performance, flexibility, and efficiency than Persistent Disk for workloads that need tuning. With Hyperdisk Balanced, IOPS and throughput can be provisioned separately and changed dynamically, which lets one volume be tuned to a wide variety of workloads. This separate control is useful when a database needs high IOPS but only moderate throughput, or when an application needs high throughput without paying for maximum IOPS.
Hyperdisk Balanced High Availability adds synchronous replication between two zones in the same region, similar to Regional Persistent Disk. Data written to the volume is replicated synchronously to the second zone, so the volume remains available if a zone outage occurs. To reduce cost and disk management complexity, capacity and performance can be purchased in advance with Hyperdisk Storage Pools.
Snapshots capture the point-in-time state of Persistent Disk volumes and protect against data loss from user error, deletion, or corruption. Snapshots are incremental by default, so only changed blocks are stored, which uses less storage space and saves money. Snapshots are stored redundantly in multiple regions with automatic checksums to ensure integrity, and the Cloud Storage location for snapshots can be configured.
Snapshots take only minutes to create even when the source disk is attached to a running instance. Snapshots can be copied across regions and used to recreate persistent disks in another region for disaster recovery. For managed protection, Backup and DR Service can create, store, and manage backups of Compute Engine VMs in their original, application-readable format for direct restore to production.
Database and shared-file workloads add availability and sharing requirements beyond a single VM disk. A self-managed third-party database on Compute Engine needs cross-zone failover logic, such as an observer instance to detect primary failure, failover rules that avoid split-brain, and an internal load balancer so applications connect to the new primary without changing addresses. Shared data across many VMs, such as configuration files, tools, and centralized logs, can use a Filestore regional instance that is replicated synchronously across three zones and can be mounted on multiple VMs.
| Option | Replication scope | Choose when |
|---|---|---|
| Zonal Persistent Disk | Single zone with built-in redundancy | The workload needs low-cost durable boot or data disks and can be rebuilt or restored in another zone |
| Regional Persistent Disk | Synchronous replication across two zones | The workload, such as a self-managed database, must stay attached and available through a zonal outage |
| Hyperdisk Balanced | Single-zone volume with separate IOPS and throughput controls | The workload needs tunable performance or dynamic resizing beyond what size-based scaling provides |
| Hyperdisk Balanced High Availability | Synchronous replication across two zones | The workload needs both tunable performance and survival of a zonal outage without restore |
Exam tip: Use Regional Persistent Disk or Hyperdisk Balanced High Availability to survive a zonal outage without restore; use snapshots to recover from deletion, corruption, or user error.
Optimizing performance with Google Cloud Hyperdisk means scaling speed controls independently from stored data size. Google Cloud Hyperdisk (Hyperdisk) is durable block storage for Compute Engine (opens in a new tab) instances that stays preserved even if the instance is stopped. Hyperdisk volumes offer configurable performance and can be resized dynamically without replacing the disk, and Hyperdisk Storage Pools can lower cost by buying capacity and performance in advance.
Decoupled performance means IOPS, throughput, and storage capacity can be scaled independently on a Hyperdisk volume. With Hyperdisk Balanced, IOPS and throughput can be provisioned separately and changed dynamically, so the volume can be tuned to match the workload instead of buying a larger disk to get more speed. This is different from Persistent Disk, where performance scales with the size of the volume.
Use Hyperdisk when a workload needs high or uneven performance on a small disk, or when performance needs change over time without changing stored data size. For example, a small database that needs high IOPS can provision high IOPS on a small Hyperdisk Balanced volume, while the same result with Persistent Disk would require provisioning a much larger disk.
Hyperdisk types are identified by reference names used in the Google Cloud CLI, REST, and client libraries, and each type has its own size range and performance controls. The actual IOPS and throughput that can be set also depend on the volume size, and the total size of all Hyperdisk volumes attached to an instance cannot exceed that instance's Hyperdisk size and attachment limits. The following comparison shows the provisioned controls for each type.
| Hyperdisk type | Size range | Performance control |
|---|---|---|
Hyperdisk Balanced (hyperdisk-balanced) | 4 GiB to 64 TiB, default 100 GiB | Provision IOPS up to 160,000 and throughput up to 2,400 MiB/s separately |
Hyperdisk Balanced High Availability (hyperdisk-balanced-high-availability) | 4 GiB to 64 TiB, default 100 GiB | Provision IOPS up to 100,000 and throughput up to 2,400 MiB/s; data is replicated synchronously across two zones in the same region |
Hyperdisk Extreme (hyperdisk-extreme) | 64 GiB to 64 TiB, default 1 TiB | Provision IOPS up to 350,000; throughput is not provisioned and equals 250 MiB/s per 1,000 IOPS, up to 5,000 MiB/s |
Hyperdisk Throughput (hyperdisk-throughput) | 2 TiB to 32 TiB, default 2 TiB | Provision throughput up to 2,400 MiB/s; IOPS is not provisioned and equals 4 IOPS per MiB/s, up to 9,600 IOPS |
Hyperdisk ML (hyperdisk-ml) | 4 GiB to 64 TiB, default 100 GiB | Provision throughput up to 2,097,152 MiB/s; IOPS is not provisioned and equals 16 IOPS per MiB/s, up to 33,554,432 IOPS |
Hyperdisk Balanced fits a wide variety of general workloads because both speed settings move independently. Hyperdisk Extreme fits workloads that need very high IOPS, while Hyperdisk Throughput and Hyperdisk ML fit workloads that need high sequential throughput and receive IOPS automatically from provisioned throughput. Hyperdisk Balanced High Availability fits workloads that must stay available during a zone outage because writes flow synchronously to both zones.
Exam tip: Choose Hyperdisk Balanced when IOPS and throughput must be set separately; choose Persistent Disk when the machine series does not support Hyperdisk.
High availability with Hyperdisk comes from synchronous replication between two zones in the same region. Hyperdisk Balanced High Availability and Regional Persistent Disk both protect against a zone outage this way, so a backup instance can continue using the replicated data if the primary zone fails. This failover without restore is something zonal disks alone cannot provide.
Sharing and platform fit add two more decision points. When a Hyperdisk Balanced volume is created, multi-writer mode can optionally be enabled so multiple instances access the disk concurrently. Persistent Disk is fast, durable block storage that is also preserved when the instance is stopped, and it remains the choice when durable storage is needed for a machine series that does not support Hyperdisk.
A zonal Persistent Disk keeps one copy of data in one zone and requires the virtual machine to run in that same zone, while a regional Persistent Disk keeps two synchronously replicated copies in two zones of the same region with one replica in the virtual machine zone. Regional Persistent Disk remains available through loss of one zone and provides lower Recovery Point Objective (RPO) and Recovery Time Objective (RTO) than snapshot restore, while zonal disk recovery from zone loss depends on snapshots.
Choose regional Persistent Disk when redundancy across zones matters more than maximum write performance, such as for databases and high-availability services. Choose zonal Persistent Disk when cost, broader machine-type support, and simpler operation matter more, and when snapshot-based backup meets recovery needs.
Hyperdisk Balanced lets IOPS and throughput be provisioned separately and changed dynamically, while Persistent Disk performance scales with volume size. For example, a small database that needs high IOPS can provision high IOPS on a small Hyperdisk Balanced volume, while the same result with Persistent Disk would require provisioning a much larger disk.
Regional Persistent Disk attaches only to virtual machines that use E2, N1, N2, and N2D machine types, cannot be created from an operating system image or from a disk created from an operating system image, and can only be resized larger. A regional Standard Persistent Disk has a minimum size of 200 GiB, and a regional Balanced Persistent Disk in read-only mode can attach to a maximum of 10 virtual machine instances.
An organization is deploying a transaction processing database on a Google Cloud Compute Engine virtual machine (VM). The workload has the following architectural requirements:
The engineering team wants to meet the performance targets cost-effectively without provisioning unneeded storage capacity.
Which storage type should the team select?
Last updated