Block storage — Hyperdisk, Persistent Disk, Local SSD, and the one that loses your data on a stop
The Compute Engine block-storage families and the distinction that decides which one you want: Hyperdisk performance is configured independently of size, Persistent Disk performance scales with capacity, Local SSD is physically attached and ephemeral. Plus the zonal/regional split and what the page says about resizing and type changes.
markdownreferencev1WGFX-1ZR0stage: designtrust: officialfamily: storageservice: compute-engineinterface: anyhyperdiskhyperdisk balancedhyperdisk balanced high availabilityhyperdisk extremehyperdisk mlhyperdisk throughputpersistent diskregional persistent disklocal ssddisk typezonal disk
Block storage on Compute Engine
1. Three families, and what actually separates them
| Family | Types the page names | The distinguishing property |
|---|---|---|
| Hyperdisk | Balanced, Balanced High Availability, Extreme, ML, Throughput | performance is configured independently of size |
| Persistent Disk | Standard, Extreme, Regional | performance scales with capacity |
| Local SSD | — | "physically attached to the server that hosts your instance", and ephemeral |
That middle column is the whole decision. On Persistent Disk, buying more IOPS means buying more gigabytes; on Hyperdisk it does not. A migration sizing exercise that was done against a per-capacity performance model and lands on Hyperdisk will over-provision capacity, sometimes by a lot. That is a saving worth finding, and it is the single most common Google Cloud block-storage estimate error.
2. Local SSD — the one to be explicit about
Data on Local SSD is lost if the VM stops, crashes or fails. Not "at risk": lost. It is a cache, a scratch area or a shard of something reconstructible, and nothing else. A lift-and-shift that maps an on-premises local disk to Local SSD because the name matches has moved a durable disk onto ephemeral media.
3. Zonal and regional
- a zonal disk is a zonal resource, and, verbatim: *"To attach a zonal persistent disk to an instance,
both resources must be in the same zone."*
- Regional Persistent Disk gives "cross-zonal synchronous disk replication";
- Hyperdisk Balanced High Availability is the Hyperdisk family's regional-replication option.
So there are two regional block-storage answers, one per family, and they are not the same product. Pick the one that matches the family you chose in §1 rather than reaching for Regional PD by habit.
4. What can change after creation
The page names "Modify Hyperdisk size and performance" and "Change the disk type" as capabilities. So a type change is supported — but read the specific type pair's own page before planning one, because this overview does not state which pairs are allowed, nor whether the change is online.
What cannot change is the zone. A zonal disk does not move zones; it is copied
(reference/overview/regions-zones-and-locations.md §3). In Terraform that shows up as a replace, which
means a destroy of the data unless a snapshot stands behind it — see guides/iac/terraform-first-apply.md
step 6.
5. An absence, recorded
The page does not address attaching a disk to multiple instances. Multi-writer attachment is a real feature on some Google Cloud disk types, and this document does not say which, because the page read here does not. Establish it from the specific disk type's page before designing a shared-disk cluster.
Next: reference/storage/filestore.md for the shared-filesystem answer, which is usually the right one when a
shared block device was the question.