Google Cloud
Block & File Storage — Persistent Disk & Filestore
Choose between block, file and object storage, and size a disk for the throughput it actually needs.
Three shapes, and the choice is usually decided by how many machines need the data.
Persistent Disk is a drive in the building that follows you between desks. Local SSD is a notepad on the desk itself — faster to reach, gone when you move. Filestore is the shared drive the whole floor mounts.
Key Concepts
1
Persistent Disk block attached to a VM. Network-backed.
Local SSD block physically attached. Fast, EPHEMERAL.
Filestore file managed NFS, many clients
Cloud Storage object HTTP API, not a filesystem2
Persistent Disk is network storage, which is why it has a property that surprises people: it can be attached read-only to many VMs at once, and it survives the VM entirely.
3
The types.
pd-standard HDD, cheap, sequential throughput
pd-balanced SSD, the sensible default
pd-ssd higher IOPS per GB
pd-extreme provision IOPS independently
hyperdisk newest: size, IOPS and throughput all separate4
Performance scales with size on most types. A small pd-balanced disk is slow not because it is balanced but because it is small — which is why the fix for a slow disk is often to grow it rather than change type. Hyperdisk breaks that coupling.
5
Local SSD is the one with a sharp edge. It is physically attached and very fast, and the data is gone when the VM stops, is preempted or live-migrates away. It is a scratch disk and a cache, never a store.
6
Disks can be resized and types changed without downtime — grow the disk, then extend the filesystem. Shrinking is not possible.
7
Snapshots are incremental and global. A snapshot taken in one region can create a disk in another, which makes them a genuine DR tool rather than only a backup.
8
Filestore is managed NFS for applications that need a shared POSIX filesystem — media processing, legacy lift-and-shift, GKE workloads needing ReadWriteMany. Tiers run from Basic to Enterprise, which adds regional availability and snapshots.
9
Regional Persistent Disk replicates synchronously across two zones, so a zone failure does not lose the disk — the right choice for a self-managed database that must survive one.
10
What the interviewer is probing.1. "What is the risk with Local SSD?" Probing: ephemerality. Stalls: "It is just fast
storage." Moves up: the data is gone when the VM stops, is preempted or live-migrates — it is
scratch space, never a store.
11
2. "Your pd-balanced disk is slow. What is the likely fix?" Probing: the size coupling.
Stalls: "Change to pd-ssd." Moves up: performance scales with size on most types, so growing the
disk is often cheaper and faster than changing type.
12
3. "When would you use a regional Persistent Disk?" Probing: zone failure. Stalls: "For more
space." Moves up: when a self-managed database must survive a zone loss — it replicates
synchronously across two zones.
13
4. "Can a Persistent Disk be attached to several VMs?" Probing: the network-storage property.
Stalls: "No, one at a time." Moves up: read-only to many at once, which surprises people coming
from other clouds — and it survives the VM entirely.